Description:FORM 2

THE PATENTS ACT, 1970
(39 of 1970)
&
THE PATENT RULES, 2003

COMPLETE SPECIFICATION
(See Section 10 and Rule 13)

Title of invention:
SYSTEM AND METHOD FOR IMPROVING GROUND TRUTH LABEL QUALITY FOR LULC CLASSIFICATION

Applicant
Tata Consultancy Services Limited
A company Incorporated in India under the Companies Act, 1956
Having address:
Nirmal Building, 9th Floor,
Nariman Point, Mumbai 400021,
Maharashtra, India

The following specification particularly describes the invention and the manner in which it is to be performed.

1

TECHNICAL FIELD
The disclosure herein generally relates to the field of improving ground truth label quality using domain knowledge, and more specifically to a system and method for improving ground truth label quality for Land Use/Land Cover (LULC) classification by fusing the predicted domain knowledge, which is in the form of LULC maps.

BACKGROUND
Land Use/Land Cover (LULC) classification is an important and crucial task for applications such as forest resource management, urban planning and change detection. In the past, many approaches have been proposed for LULC classification including manual ground survey, SVM and fuzzy logic based techniques, K-means clustering techniques, and Random Forest methods. With the rise of deep learning, many Convolution Neural Network (CNN) based approaches have been applied successfully to solve LULC classification problem. LULC classification using multiple modalities have shown better results than using a single modality. Especially, fusing SAR and multi-spectral data has the benefit of combining unique features to distinguish various class spectral characteristics.
Recently, widespread deployment of remote sensing sensors has caused an influx of large-volume data. Getting high-quality and high-resolution LULC ground truth labels is time and resource-consuming. On the other hand, obtaining incomplete and inaccurate labels are easy, which could be used in a weakly supervised learning paradigm. Some of the provided ground truth data could be imprecise or noisy and inexact or low-resolution. Thus, the challenge is to develop a deep learning model that can predict accurate high-resolution LULC output despite using the noisy low-resolution labels during training.

SUMMARY
Embodiments of the disclosure present technological improvements as solutions to one or more of the above-mentioned technical problems recognized by the inventors in conventional systems. For example, in one embodiment, a system and method for improving ground truth label quality for Land Use/Land Cover (LULC) classification by fusing the predicted domain knowledge, which is in the form of LULC maps is provided.
In one aspect, a processor-implemented method for improving ground truth label quality for Land Use/Land Cover (LULC) classification by fusing the predicted domain knowledge, which is in the form of LULC maps. The processor-implemented method comprising receiving a dual polarized sentinel-1 synthetic aperture radar (SAR) image, a sentinel-2 multispectral (MS) image, and low-resolution Moderate Resolution Imaging Spectroradiometer (MODIS) labels as input, generating a normalized difference vegetation index (NDVI) map, and a normalized difference water index (NDWI) map based on the sentinel-2 multispectral (MS) image, processing, via one or more hardware processors, the received low-resolution MODIS labels to refine one or more labels using the generated normalized difference vegetation index (NDVI) map, and the normalized difference water index (NDWI) map, redistributing one or more pixels of the processed low-resolution MODIS labels to a predefined classes based on a NDVI map and a NDWI map to refine the processed low-resolution MODIS labels and training a Convolutional Neural Network (CNN) model with the dual polarized sentinel-1 synthetic aperture radar (SAR) image, and sentinel-2 multispectral (MS) image using the refined low-resolution MODIS labels as a ground truth to generate land Use /Land Cover (LULC) map.
In another aspect, a system for improving ground truth label quality for Land Use/Land Cover (LULC) classification by fusing the predicted domain knowledge, which is in the form of LULC maps is provided. The system includes an input/output interface configured to receive a dual polarized sentinel-1 synthetic aperture radar (SAR) image, a sentinel-2 multispectral (MS) image, and low-resolution Moderate Resolution Imaging Spectroradiometer (MODIS) labels as input one or more hardware processors and at least one memory storing a plurality of instructions, wherein the one or more hardware processors are configured to execute the plurality of instructions stored in the at least one memory.
Further, the system is configured to generate a normalized difference vegetation index (NDVI) map, and a normalized difference water index (NDWI) map based on the sentinel-2 multispectral (MS) image, process the received low-resolution MODIS labels to remove one or more noises using the generated normalized difference vegetation index (NDVI) map, and the normalized difference water index (NDWI) map, redistribute one or more pixels of the processed low-resolution MODIS labels to a predefined classes based on a NDVI map and a NDWI map to refine the processed low-resolution MODIS labels, and train a Convolutional Neural Network (CNN) model with the dual polarized sentinel-1 synthetic aperture radar (SAR) image, and sentinel-2 multispectral (MS) image using the refined low-resolution MODIS labels as a ground truth to generate land Use /Land Cover (LULC) map, wherein the CNN model comprising one or more encoder blocks and one or more decoder blocks.
In yet another aspect, one or more non-transitory machine-readable information storage mediums are provided comprising one or more instructions, which when executed by one or more hardware processors causes a method for improving ground truth label quality for Land Use/Land Cover (LULC) classification by fusing the predicted domain knowledge, which is in the form of LULC maps. The method comprising receiving a dual polarized sentinel-1 synthetic aperture radar (SAR) image, a sentinel-2 multispectral (MS) image, and low-resolution Moderate Resolution Imaging Spectroradiometer (MODIS) labels as input, generating a normalized difference vegetation index (NDVI) map, and a normalized difference water index (NDWI) map based on the sentinel-2 multispectral (MS) image, processing, via one or more hardware processors, the received low-resolution MODIS labels to remove one or more noises using the generated normalized difference vegetation index (NDVI) map, and the normalized difference water index (NDWI) map, redistributing one or more pixels of the processed low-resolution MODIS labels to a predefined classes based on a NDVI map and a NDWI map to refine the processed low-resolution MODIS labels and training a Convolutional Neural Network (CNN) model with the dual polarized sentinel-1 synthetic aperture radar (SAR) image, and sentinel-2 multispectral (MS) image using the refined low-resolution MODIS labels as a ground truth to generate land Use /Land Cover (LULC) map.
It is to be understood that the foregoing general descriptions and the following detailed description are exemplary and explanatory only and are not restrictive of the invention, as claimed.

BRIEF DESCRIPTION OF THE DRAWINGS
The accompanying drawings, which are incorporated in and constitute a part of this disclosure, illustrate exemplary embodiments and, together with the description, serve to explain the disclosed principles:
FIG. 1 illustrates a block diagram of an exemplary system for improving ground truth label quality for Land Use/Land Cover (LULC) classification by fusing the predicted domain knowledge, which is in the form of LULC maps, in accordance with some embodiments of the present disclosure.
FIG. 2 is a block diagram to illustrate an overview of a Convolutional Neural Network architecture, in accordance with some embodiments of the present disclosure.
FIG. 3 is a flow diagram to illustrate a processor-implemented method for improving ground truth label quality for Land Use/Land Cover (LULC) classification by fusing the predicted domain knowledge, which is in the form of LULC maps, in accordance with some embodiments of the present disclosure.

DETAILED DESCRIPTION OF EMBODIMENTS
Exemplary embodiments are described with reference to the accompanying drawings. In the figures, the left-most digit(s) of a reference number identifies the figure in which the reference number first appears. Wherever convenient, the same reference numbers are used throughout the drawings to refer to the same or like parts. While examples and features of disclosed principles are described herein, modifications, adaptations, and other implementations are possible without departing from the scope of the disclosed embodiments.
Land Use / Land Cover (LULC) classification is an essential and crucial task for applications such as forest resource management, urban planning, and change detection. In the past, many approaches have been proposed for LULC classification, including manual ground survey, SVM and fuzzy logic-based techniques, K-means clustering techniques, and Random Forest methods. With the rise of deep learning, many CNN-based approaches have been applied successfully to solve the LULC classification problem. LULC classification using multiple modalities has been shown to perform better than using a single modality. Especially, fusing SAR and multi-spectral data has the benefit of combining unique features to distinguish various spectral characteristics.
The widespread deployment of remote sensing sensors has recently caused an influx of large-volume data; getting high-quality and high-resolution Land Use/Land Cover (LULC) ground truth labels is time and resource-consuming. On the other hand, obtaining incomplete and inaccurate labels are easy, which could be used in a weakly supervised learning paradigm. Some of the provided ground truth data could be imprecise or noisy and inexact, or low-resolution in this setting. Thus, the challenge is to develop a deep learning model that can predict accurate high-resolution LULC output despite using noisy low-resolution labels during training.
Existing trained CNN models utilizing symmetric cross entropy for LULC classification are using noisy low-resolution labels. Similarly, some models have trained a UNet and Deeplab-V3+ model on noisy labels. However, results of these approaches are often inaccurate compared to the expected high-resolution output. Hence, the current weakly supervised methods are far from being perfect and inadequate for real-time analysis.
Embodiments herein proposing a system and method to bridge this gap to generate high-resolution and highly accurate LULC output by using only the low-resolution noisy labels during training. A technique is proposed to refine the low-resolution noisy MODIS labels with a normalized difference vegetation index (NDVI) and a normalized difference water index (NDWI) metrics. Then, the refined labels are used to train a Convolutional Neural Network (CNN) model for improved LULC classification.
Referring now to the drawings, and more particularly to FIG. 1 through FIG. 3, where similar reference characters denote corresponding features consistently throughout the figures, there are shown preferred embodiments and these embodiments are described in the context of the following exemplary system and/or method.
FIG. 1 illustrates a block diagram of a system (100) for improving ground truth label quality for Land Use/Land Cover (LULC) classification by fusing the predicted domain knowledge, which is in the form of LULC maps. Although the present disclosure is explained considering that the system (100) is implemented on a server, it may be understood that the system (100) may comprise one or more computing devices (102), such as a laptop computer, a desktop computer, a notebook, a workstation, a cloud-based computing environment and the like. It will be understood that the system (100) may be accessed through one or more input/output interfaces 104-1, 104-2... 104-N, collectively referred to as I/O interface (104). Examples of the I/O interface (104) may include, but are not limited to, a user interface, a portable computer, a personal digital assistant, a handheld device, a smartphone, a tablet computer, a workstation, and the like. The I/O interface (104) are communicatively coupled to the system (100) through a network (106).
In an embodiment, the network (106) may be a wireless or a wired network, or a combination thereof. In an example, the network (106) can be implemented as a computer network, as one of the different types of networks, such as virtual private network (VPN), intranet, local area network (LAN), wide area network (WAN), the internet, and such. The network (106) may either be a dedicated network or a shared network, which represents an association of the different types of networks that use a variety of protocols, for example, Hypertext Transfer Protocol (HTTP), Transmission Control Protocol/Internet Protocol (TCP/IP), and Wireless Application Protocol (WAP), to communicate with each other. Further, the network (106) may include a variety of network devices, including routers, bridges, servers, computing devices, storage devices. The network devices within the network (106) may interact with the system (100) through communication links.
The system (100) supports various connectivity options such as BLUETOOTH®, USB, ZigBee, and other cellular services. The network environment enables connection of various components of the system (100) using any communication link including Internet, WAN, MAN, and so on. In an exemplary embodiment, the system (100) is implemented to operate as a stand-alone device. In another embodiment, the system (100) may be implemented to work as a loosely coupled device to a smart computing environment. Further, the system (100) comprises at least one memory with a plurality of instructions, one or more databases (112), and one or more hardware processors (108) which are communicatively coupled with the at least one memory (110) to execute a plurality of modules therein. The components and functionalities of the system (100) are described further in detail.
Herein, the one or more I/O interfaces (104) are configured to receive a dual polarized sentinel-1 synthetic aperture radar (SAR) image, a sentinel-2 multispectral (MS) image, and low-resolution Moderate Resolution Imaging Spectroradiometer (MODIS) labels as input. The Sentinel-1 is a phase-preserving dual polarization SAR arrangement. It can transmit a signal in either horizontal (H) or vertical (V) polarization, and then receive in both H and V polarizations. Dual polarization Level-1 Single Look Complex (SLC) products contain complex values. In addition to the backscatter intensity that can be measured from each single polarization, the inter-channel phase information allows to perform enhanced analysis of backscattering properties.
A multi-spectral image is a collection of several monochrome images of the same scene, each of them taken with a different sensor. Each image is referred to as a band. A well-known multi-spectral (or multi-band image) is a RGB color image, consisting of a red, a green and a blue image, each of them taken with a sensor sensitive to a different wavelength. In image processing, multi-spectral images are most commonly used for remote sensing applications. Satellites usually take several images from frequency bands in the visual and non-visual range. Sentinel-2 provides spectral bands in a wide range of the electromagnetic spectrum – including visible, NIR, short wave infrared, and four red edge bands – which are freely available data that can be very effective for LULC classification.
In one embodiment, the system (100) is configured to generate a normalized difference vegetation index (NDVI) map, and a normalized difference water index (NDWI) map based on the sentinel-2 multispectral (MS) image. The NDVI map quantifies vegetation by measuring the difference between near-infrared (NIR) (which vegetation strongly reflects) and red light (which vegetation absorbs). NDVI always ranges from -1 to +1. But there isn’t a distinct boundary for each type of land cover. When there are negative values, it’s highly likely that it’s water. On the other hand, if the NDVI value close to +1, there’s a high possibility that it’s dense green leaves. Healthy vegetation (chlorophyll) reflects more near-infrared (NIR) and green light compared to other wavelengths, but it absorbs more red and blue light. The NDVI uses the NIR and red channels in its formula as:
NDVI = ((NIR-Red))/((NIR+Red)) (1)
Further, the NDWI map is used to highlight open water features in a satellite image, allowing a water body to stand out against the soil and vegetation. Taking advantage of the NIR and visible green spectral bands, the NDWI is capable of enhancing the water bodies in a satellite image. The downside of the index is that it is sensitive to build structures, which can lead to overestimation of water bodies. The NDWI is calculated using the green-NIR (visible green and near-infrared) combination, which allows it to detect subtle changes in water content of the water bodies as:
NDWI = ((Green-NIR))/((Green+NIR)) (2)
The visible green wavelengths maximize the typical reflectance of the water surface. The near-infrared wavelengths maximize the high reflectance of terrestrial vegetation and soil features, while minimizing the low reflectance of water features. The result of the NDWI equation is positive values for water features and negative ones (or zero) for soil and terrestrial vegetation. Beyond the visible spectrum towards the infrared, water reflects almost no light. The NDWI makes use of this property to successfully outline water bodies on the map and monitor water’s turbidity.
In another embodiment, the system (100) is configured to process the received low-resolution MODIS labels to refine one or more labels using the NDVI, and the NDWI indices. Further, the one or more pixels of the processed low-resolution MODIS labels are redistributed to other predefined classes based on the NDVI and NDWI indices. The classes are redistributed based on the NDWI and NDVI values. For NDWI values beyond 0, the pixels are mapped to water class. Similarly, NDWI between -0.6 and 0, are mapped to Forest class. NDWI values between -0.6 and -0.5 are mapped to Wetlands and -0.5 to -0.4 are mapped to Grasslands. For NDWI values between -0.3 and 0, and NDVI values between 0.1 and 0.2 are mapped to Croplands.
In order to improve the label accuracy, the labels are reassigned based on the NDVI and the NDWI metrics. For pixels with NDWI values above 0, irrespective of NDVI values, are relabeled as water class. Similarly, NDWI values less than -0.6 is reassigned to forest class. The pixels with NDWI values between -0.4 and -0.5 are relabeled as grasslands, and -0.5 and -0.6 as wetlands class. Also, pixels with NDWI values between -0.4 and -0.3, with NDVI values less than 0.3 is reassigned to croplands class. With same range of NDWI values, for NDVI values between 0.3 and 0.5, is relabeled to shrublands class, and NDVI values greater than 0.5 is relabeled to grasslands class. Similarly, for NDWI values between 0 and -0.3, and NDVI values less than 0.1 is reassigned to urban class. For same range of NDWI values, and NDVI between 0.1 and 0.2 the pixels are relabeled to croplands class and for NDVI more than 0.2 is relabeled as barren class.
In another embodiment, the system (100) is configured to train a Convolutional Neural Network (CNN) model with the dual polarized sentinel-1 synthetic aperture radar (SAR) image, and sentinel-2 multispectral (MS) image using the refined low-resolution MODIS labels as a ground truth in a weakly supervised fashion to generate land Use /Land Cover (LULC) map. The CNN model comprising one or more encoder blocks and one or more decoder blocks as shown in FIG. 2. The one or more encoder blocks comprising a predefined series of a convolution layer, a maxpooling layer and a leakyReLU activation function. The one or more decoder blocks comprising a deconvolutional layer with a skip connection from the encoder to generate LULC map output in the same resolution as input.
Finally, the refined labels are used as ground truth to train CNN model with quadruplet inputs such as 3-channel multispectral data (red, blue, NIR bands), 2-channel SAR data, NDWI map, and NDVI map. All four inputs are concatenated in the channel dimension to form a seven-channel input to the CNN model. Further, the categorical cross entropy (CCE) loss is computed between the predicted LULC map and the redefined MODIS label maps to train the CNN model.
In one illustration, the CNN model is trained for 200 epochs with a learning rate of 10-4 and an Adam optimizer with 0.9 momentum. The system is configured to address the class imbalance problem by weighing the class-wise loss values with [40, 30, 40, 20, 40, 100, 30, 15] weight vector. The class weights were assigned based on the frequency of a particular class in the training dataset, widespread class is assigned low weight and vice-versa. The exact weight vector is finalized through an extensive grid search. To train and test the CNN model, the dataset consists of 984 training patches and 5128 testing patches. Each patch consists of a dual-polarized Sentinel-1 SAR image, a 13-band multispectral Sentinel-2 image, a 500m low-resolution MODIS land cover map, and a 10m high-resolution land cover map. Empirically, it is found that using only red, blue, and NIR bands in the Sentinel-2 multi-spectral image results in high accuracy; thus, only those three bands are used in IMS as input. Also, as mentioned earlier, high-resolution land cover maps do not use in any way during the training.
Referring FIG. 3, a flow chart (300), illustrating a processor-implemented method for improving ground truth label quality for Land Use/Land Cover (LULC) classification by fusing the predicted domain knowledge, which is in the form of LULC maps is provided. Initially, at step (302), receiving a dual polarized sentinel-1 synthetic aperture radar (SAR) image, a sentinel-2 multispectral (MS) image, and low-resolution Moderate Resolution Imaging Spectroradiometer (MODIS) labels as input.
At the next step (304), generating a normalized difference vegetation index (NDVI) map, and a normalized difference water index (NDWI) map based on the sentinel-2 multispectral (MS) image. Wherein, red, green and NIR bands of the sentinel-2 multispectral images are used to generate the normalized difference water index (NDWI) map and the normalized difference water index (NDWI) map.
At the next step (306), processing the received low-resolution MODIS labels to remove one or more noises using the generated normalized difference vegetation index (NDVI) map, and the normalized difference water index (NDWI) map.
At the next step (308), redistributing one or more pixels of the processed low-resolution MODIS labels to a predefined classes based on a NDVI map and a NDWI map to refine the processed low-resolution MODIS labels, wherein the predefined classes are redistributed based on one or more values of the NDWI map and NDVI map.
At the next step (310), training a Convolutional Neural Network (CNN) model with the dual polarized sentinel-1 synthetic aperture radar (SAR) image, and sentinel-2 multispectral (MS) image using the refined low-resolution MODIS labels as a ground truth in a weakly supervised fashion to generate land Use /Land Cover (LULC) map, wherein the CNN model comprising one or more encoder blocks and one or more decoder blocks.
The written description describes the subject matter herein to enable any person skilled in the art to make and use the embodiments. The scope of the subject matter embodiments is defined by the claims and may include other modifications that occur to those skilled in the art. Such other modifications are intended to be within the scope of the claims if they have similar elements that do not differ from the literal language of the claims or if they include equivalent elements with insubstantial differences from the literal language of the claims.
The embodiments of present disclosure herein address the problem of fusing synthetic aperture radar (SAR) and multispectral image to generate a precise Land Use/Land Cover (LULC) classification in a weakly supervised setting is a challenging yet important problem. The inaccurate, noisy, and inexact ground truth labels pose difficulty to train any machine learning models. Embodiments herein proposing a system and method to bridge this gap to generate high-resolution and highly accurate LULC output by using only the low-resolution noisy labels during training. A technique is proposed to refine the low-resolution noisy MODIS labels with normalized difference vegetation index (NDVI) and normalized difference water index (NDWI) metrics. Then, the refined labels are used to train a Convolutional Neural Network (CNN) model to improve LULC classification.
It is to be understood that the scope of the protection is extended to such a program and in addition to a computer-readable means having a message therein; such computer-readable storage means contain program-code means for implementation of one or more steps of the method, when the program runs on a server or mobile device or any suitable programmable device. The hardware device can be any kind of device which can be programmed including e.g., any kind of computer like a server or a personal computer, or the like, or any combination thereof. The device may also include means which could be e.g., hardware means like e.g., an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a combination of hardware and software means, e.g., an ASIC and an FPGA, or at least one microprocessor and at least one memory with software modules located therein. Thus, the means can include both hardware means, and software means. The method embodiments described herein could be implemented in hardware and software. The device may also include software means. Alternatively, the embodiments may be implemented on different hardware devices, e.g., using a plurality of CPUs.
The embodiments herein can comprise hardware and software elements. The embodiments that are implemented in software include but are not limited to, firmware, resident software, microcode, etc. The functions performed by various modules described herein may be implemented in other modules or combinations of other modules. For the purposes of this description, a computer-usable or computer readable medium can be any apparatus that can comprise, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device.
The illustrated steps are set out to explain the exemplary embodiments shown, and it should be anticipated that ongoing technological development will change the manner in which particular functions are performed. These examples are presented herein for purposes of illustration, and not limitation. Further, the boundaries of the functional building blocks have been arbitrarily defined herein for the convenience of the description. Alternative boundaries can be defined so long as the specified functions and relationships thereof are appropriately performed. Alternatives (including equivalents, extensions, variations, deviations, etc., of those described herein) will be apparent to persons skilled in the relevant art(s) based on the teachings contained herein. Such alternatives fall within the scope of the disclosed embodiments. Also, the words “comprising,” “having,” “containing,” and “including,” and other similar forms are intended to be equivalent in meaning and be open ended in that an item or items following any one of these words is not meant to be an exhaustive listing of such item or items or meant to be limited to only the listed item or items. It must also be noted that as used herein and in the appended claims, the singular forms “a,” “an,” and “the” include plural references unless the context clearly dictates otherwise.
Furthermore, one or more computer-readable storage media may be utilized in implementing embodiments consistent with the present disclosure. A computer-readable storage medium refers to any type of physical memory on which information or data readable by a processor may be stored. Thus, a computer-readable storage medium may store instructions for execution by one or more processors, including instructions for causing the processor(s) to perform steps or stages consistent with the embodiments described herein. The term “computer-readable medium” should be understood to include tangible items and exclude carrier waves and transient signals, i.e., be non-transitory. Examples include random access memory (RAM), read-only memory (ROM), volatile memory, nonvolatile memory, hard drives, CD ROMs, DVDs, flash drives, disks, and any other known physical storage media.
It is intended that the disclosure and examples be considered as exemplary only, with a true scope of disclosed embodiments being indicated by the following claims.
, Claims:

1. A processor-implemented method (300) for training a Convolutional Neural Network (CNN) model with noisy low-resolution labels to generate a land Use /Land Cover (LULC) map comprising steps of:
receiving (302), via an input/output interface, a dual polarized sentinel-1 synthetic aperture radar (SAR) image, a sentinel-2 multispectral (MS) image, and low-resolution Moderate Resolution Imaging Spectroradiometer (MODIS) labels as inputs;
generating (304), via one or more hardware processors, a normalized difference vegetation index (NDVI) map, and a normalized difference water index (NDWI) map based on the sentinel-2 multispectral (MS) image;
processing (306), via one or more hardware processors, the received low-resolution MODIS labels to remove one or more noises using the generated normalized difference vegetation index (NDVI) map, and the normalized difference water index (NDWI) map;
redistributing (308), via the one or more hardware processors, one or more pixels of the processed low-resolution MODIS labels to a plurality of predefined classes based on the NDVI map and the NDWI map to refine the processed low-resolution MODIS labels, wherein the plurality of predefined classes are redistributed based on one or more values of the NDWI map and the NDVI map; and
training (310), via the one or more hardware processors, a Convolutional Neural Network (CNN) model with the dual polarized sentinel-1 synthetic aperture radar (SAR) image, and the sentinel-2 multispectral (MS) image using the refined low-resolution MODIS labels as a ground truth in a weakly supervised fashion to generate the land Use /Land Cover (LULC) map.
2. The processor-implemented method (300) of claim 1, wherein red, green and near-infrared (NIR) bands of the sentinel-2 multispectral image are used to generate the normalized difference water index (NDWI) map and the normalized difference water index (NDWI) map.
3. The processor-implemented method (300) of claim 1, wherein the one or more encoder blocks comprising a predefined series of a convolution layer, a maxpooling layer and a leakyReLU activation function.
4. The processor-implemented method (300) of claim 1, wherein the one or more decoder blocks comprising a deconvolutional layer with a skip connection from the encoder.
5. A system (100) for training a Convolutional Neural Network (CNN) model with noisy low-resolution labels to generate a land Use /Land Cover (LULC) map comprising:
an input/output interface (104) to receive a dual polarized sentinel-1 synthetic aperture radar (SAR) image, a sentinel-2 multispectral (MS) image, and low-resolution Moderate Resolution Imaging Spectroradiometer (MODIS) labels as inputs;
a memory (110) in communication with the one or more hardware processors (108), wherein the one or more hardware processors (108) are configured to execute programmed instructions stored in the memory (110) to:
generate a normalized difference vegetation index (NDVI) map, and a normalized difference water index (NDWI) map based on the sentinel-2 multispectral (MS) image;
process the received low-resolution MODIS labels to remove one or more noises using the generated normalized difference vegetation index (NDVI) map, and the normalized difference water index (NDWI) map;
redistribute one or more pixels of the processed low-resolution MODIS labels to a plurality of predefined classes based on the NDVI map and the NDWI map to refine the processed low-resolution MODIS labels, wherein the plurality of predefined classes are redistributed based on one or more values of the NDWI map and the NDVI map; and
train a Convolutional Neural Network (CNN) model with the dual polarized sentinel-1 synthetic aperture radar (SAR) image, and the sentinel-2 multispectral (MS) image using the refined low-resolution MODIS labels as a ground truth to in a weakly supervised fashion generate land Use /Land Cover (LULC) map,.
6. The system (100) of claim 5, wherein red, green and NIR bands of the sentinel-2 multispectral images are used to generate the normalized difference vegetation index (NDVI) map and the normalized difference water index (NDWI) map.
7. The system (100) of claim 5, wherein the one or more encoder blocks comprising a predefined series of a convolution layer, a maxpooling layer and a leakyReLU activation function.
8. The system (100) of claim 5, wherein the one or more decoder blocks comprising a deconvolutional layer with a skip connection from the encoder.
9. A non-transitory computer readable medium storing one or more instructions which when executed by one or more processors on a system, cause the one or more processors to perform method comprising:
receiving, via an input/output interface, a dual polarized sentinel-1 synthetic aperture radar (SAR) image, a sentinel-2 multispectral (MS) image, and low-resolution Moderate Resolution Imaging Spectroradiometer (MODIS) labels as input;
generating, via one or more hardware processors, a normalized difference vegetation index (NDVI) map, and a normalized difference water index (NDWI) map based on the sentinel-2 multispectral (MS) image;
processing, via one or more hardware processors, the received low-resolution MODIS labels to remove one or more noises using the generated normalized difference vegetation index (NDVI) map, and the normalized difference water index (NDWI) map;
redistributing, via the one or more hardware processors, one or more pixels of the processed low-resolution MODIS labels to a plurality of predefined classes based on the NDVI map and the NDWI map to refine the processed low-resolution MODIS labels, wherein the plurality of predefined classes are redistributed based on one or more values of the NDWI map and the NDVI map; and
training, via the one or more hardware processors, a Convolutional Neural Network (CNN) model with the dual polarized sentinel-1 synthetic aperture radar (SAR) image, and the sentinel-2 multispectral (MS) image using the refined low-resolution MODIS labels as a ground truth in a weakly supervised fashion to generate the land Use /Land Cover (LULC) map.