1. A computer-implemented method for enabling adaptive learning in Liquid Neural Network (108) (LNN) based land use and land cover (LULC) models under spatial and temporal domain shifts, comprising: a. a multispectral satellite data (102) is collected from one or more remote sensing sources; b. the collected multispectral satellite data (102) is preprocessed (104) using cleaning and normalization to improve consistency and remove errors; c. spectral and spatial features (106) are extracted from the preprocessed multispectral satellite data (102); d. a Liquid Neural Network (108) (LNN) model is designed and implemented, said LNN model being configured to process the extracted spectral and spatial features (106); e. the LNN model is trained using labeled data corresponding to different land use and land cover categories; f. an adaptive learning mechanism (110) is incorporated within the LNN (108) model, said adaptive learning mechanism (110) being configured to update parameters of the LNN (108) model based on variations in input data; when g. multispectral satellite data (102) having different spatial and temporal characteristics is received, said adaptive learning mechanism (110) adjusts the LNN model to maintain classification accuracy without requiring complete retraining; h. the method enables handling of spatial variations across different geographic regions and temporal variations across different time periods to improve robustness and generalization; and i. wherein said method generates classification output in the form of labeled maps or datasets (112) representing different land use and land cover categories.

2. The method as claimed in claim 1, wherein the preprocessing (104) of the multispectral satellite data (102) comprises noise reduction and correction to improve data quality.

3. The method as claimed in claim 1, wherein the spectral and spatial features (106) include information related to reflectance, texture, and spatial patterns of land regions.

4. The method as claimed in claim 1, wherein said Liquid Neural Network (108) (LNN) model is configured to capture temporal variations present in the multispectral satellite data (102).

5. The method as claimed in claim 1, wherein said adaptive learning mechanism (110) enables continuous updating of parameters of the LNN model without requiring complete retraining of the model.

6. The method as claimed in claim 1, wherein said adaptive learning mechanism (110) is configured to respond to changes in data distribution across different geographic regions.

7. The method as claimed in claim 1, wherein said adaptive learning mechanism (110) is configured to adjust the LNN model based on seasonal or time-based variations in the input data.

8. The method as claimed in claim 1, wherein the classification output includes identification of land categories comprising forest, water, urban, and agricultural land.

9. The method as claimed in claim 1, wherein said method is implemented in a computer system for real-time or near real-time land use and land cover classification.

10. The method as claimed in claim 1, wherein the generated classification output is provided in the form of maps or structured datasets (112) for analysis.

6. DATE AND SIGNATURE Dated this on 06th April 2026 Signature Mr. Srinivas Maddipati (IN/PA 3124) Agent for applicant

Description:4. DESCRIPTION
Technical Field of the Invention

The present invention relates to the field of remote sensing and machine learning, More particularly, the invention relates to a computer-implemented method for enabling adaptive learning in Liquid Neural Network (LNN) based land use and land cover (LULC) classification models under spatial and temporal domain shifts.

Background of the Invention

Land Use and Land Cover (LULC) classification is widely used in applications such as environmental monitoring, urban planning, and agriculture management. It helps in understanding how land is being used and how it changes over time. Satellite imagery is commonly used for this purpose because it provides large-scale and continuous data.

Traditional machine learning and deep learning models have been used for LULC classification. These models are usually trained on a fixed dataset and perform well only when new data is similar to the training data. However, in real-world conditions, this assumption is often not valid.

Satellite data changes due to various factors such as seasonal variations, weather conditions, and differences in geographical regions. These changes create what is known as spatial and temporal domain shifts. As a result, the performance of existing models decreases when applied to new or unseen conditions.

Many existing solutions try to solve this problem by retraining the model with new data. This approach requires additional time, computational resources, and labelled data, which may not always be available. It also makes the system less efficient for real-time applications.

Some advanced methods attempt domain adaptation, but they are often complex and not flexible enough to handle continuous changes in data. These methods may still struggle when both spatial and temporal variations occur together.

Accordingly, there exists a need for a method that can automatically adapt to changing data conditions. Such a method should improve model performance without requiring frequent retraining and should work effectively across different locations and time periods.
Objects of the Invention

The principal object of the present invention is to provide a method for enabling adaptive learning in Liquid Neural Network (LNN)-based land use and land cover (LULC) models.

Another object of the present invention is to handle spatial and temporal domain shifts in multispectral satellite data in an efficient manner.

Another object of the present invention is to improve the accuracy and consistency of LULC classification under varying environmental and geographical conditions.

Another object of the present invention is to reduce the need for frequent retraining of models when new data is introduced.

Another object of the present invention is to enhance the robustness and generalization capability of LNN-based models.

Another object of the present invention is to provide a simple and efficient approach that can be applied in real-time remote sensing applications.

Brief Summary of the Invention

The present invention relates to a method for enabling adaptive learning in Liquid Neural Network (LNN)-based land use and land cover (LULC) models. The invention focuses on improving the performance of classification models when there are changes in input data across different locations and time periods.

In one embodiment, multispectral satellite data is collected from suitable remote sensing sources. The collected data is then subjected to preprocessing steps such as cleaning, normalization, and basic corrections to improve data quality and consistency.

After preprocessing, important features are extracted from the data. These features include spectral and spatial information that help in distinguishing different types of land use and land cover.

Designing and implementing a Liquid Neural Network architecture. The model is trained using labeled datasets so that it can learn patterns associated with various land categories.

Introduction of an adaptive learning mechanism within the LNN model. This mechanism allows the model to adjust its internal parameters when it encounters new data with different characteristics.

The method is capable of handling both spatial variations, such as changes across different geographic regions, and temporal variations, such as seasonal changes over time. This helps in maintaining consistent performance under real-world conditions.

As a result, the invention provides improved classification accuracy, better generalization, and reduced dependence on frequent retraining. The output is generated in the form of reliable classified maps or labeled datasets for practical applications.
Brief Description of the Drawings

The invention will be further understood from the following detailed description of a preferred embodiment taken in conjunction with an appended drawing, in which:

Fig. 1 illustrates a flow diagram of a method for enabling adaptive learning in Liquid Neural Network (108) (LNN) based land use and land cover (LULC) models, in accordance with an exemplary embodiment of the present invention;

Detailed Description of the Invention

It is to be understood that the present disclosure is not limited in its application to the details of construction and the arrangement of components set forth in the following description or illustrated in the drawings. The present disclosure is capable of other embodiments and of being practiced or of being carried out in various ways. In addition, it is to be understood that the phraseology and terminology used herein is for the purpose of description and should not be regarded as limiting.

The use of “including”, “comprising” or “having” and variations thereof herein is meant to encompass the items listed thereafter and equivalents thereof as well as additional items. The terms “a” and “an” herein do not denote a limitation of quantity, but rather denote the presence of at least one of the referenced items. Further, the use of terms “first”, “second”, and “third”, and the like, herein do not denote any order, quantity, or importance, but rather are used to distinguish one element from another.

According to an exemplary embodiment of the present invention, a method for enabling adaptive learning in Liquid Neural Network (LNN)-based land use and land cover (LULC) models under spatial and temporal domain shifts is disclosed. The method focuses on improving classification performance when input data varies across different regions and time periods. It provides a flexible approach that can adjust to real-world changes without affecting overall system efficiency.

In accordance with an exemplary embodiment of the present invention, multispectral satellite data is collected from one or more remote sensing sources. The data may include multiple spectral bands useful for identifying different land types. Such data enables better understanding of surface characteristics and improves classification capability.

In accordance with an exemplary embodiment of the present invention, the collected data is subjected to pre-processing including cleaning, normalization, and basic correction. This step helps in improving consistency and removing unwanted noise from the data. It ensures that the input data is suitable for further processing and reduces errors in later stages.

In accordance with an exemplary embodiment of the present invention, spectral and spatial features are extracted from the pre-processed data. These features help in distinguishing between different land use and land cover categories. The extracted features improve the ability of the model to identify patterns more accurately.

In accordance with an exemplary embodiment of the present invention, a Liquid Neural Network (LNN) model is designed and implemented. The model is capable of handling dynamic changes in input data due to its continuous learning behaviour. This makes it suitable for applications where data conditions frequently change.

In accordance with an exemplary embodiment of the present invention, the LNN model is trained using labelled datasets representing different land categories. This training enables the model to learn patterns associated with each category. Proper training helps in improving prediction accuracy and model reliability.
In accordance with an exemplary embodiment of the present invention, an adaptive learning mechanism is incorporated within the LNN model. This mechanism allows the model to update its parameters when new data is received. It ensures that the model remains effective even when data conditions change.

In accordance with an exemplary embodiment of the present invention, the adaptive learning mechanism detects variations in input data caused by spatial and temporal domain shifts. Based on these variations, the model adjusts its behaviour to maintain performance. This reduces performance drop when applied to new regions or time periods.

In accordance with an exemplary embodiment of the present invention, the method handles spatial variations such as differences between geographic regions. It also handles temporal variations such as seasonal and time-based changes. This improves the adaptability of the model in real-world scenarios.

In accordance with an exemplary embodiment of the present invention, the LNN model performs classification of the multispectral data into predefined land use and land cover categories. These categories may include forest, water, urban, and agricultural land. The classification process provides meaningful insights into land usage patterns.

In accordance with an exemplary embodiment of the present invention, the output is generated in the form of classified maps or labelled datasets. This output can be used for various applications such as monitoring and planning. It supports decision-making in environmental and urban management.

In accordance with an exemplary embodiment of the present invention, the method improves accuracy, robustness, and generalization of LULC models. It also reduces the need for frequent retraining and supports efficient real-world implementation. The invention therefore provides a practical and scalable solution for handling changing data conditions.

In references to figures, Fig. 1 illustrates the method begins with collecting multispectral satellite data (102) from one or more remote sensing sources. The collected data is then preprocessed (104) using cleaning and normalization to improve data quality. Thereafter, spectral and spatial features (106) are extracted to support classification. A Liquid Neural Network (108) (LNN) model is then designed and trained using labeled data. An adaptive learning mechanism (110) is incorporated within the LNN model to update its parameters based on variations in input data. Finally, the method generates classification output in the form of classified maps or labeled datasets (112) representing different land use and land cover categories.
, Claims:CLAIMS
We Claim:
1. A computer-implemented method for enabling adaptive learning in Liquid Neural Network (108) (LNN) based land use and land cover (LULC) models under spatial and temporal domain shifts, comprising:
a. a multispectral satellite data (102) is collected from one or more remote sensing sources;
b. the collected multispectral satellite data (102) is preprocessed (104) using cleaning and normalization to improve consistency and remove errors;
c. spectral and spatial features (106) are extracted from the preprocessed multispectral satellite data (102);
d. a Liquid Neural Network (108) (LNN) model is designed and implemented, said LNN model being configured to process the extracted spectral and spatial features (106);
e. the LNN model is trained using labeled data corresponding to different land use and land cover categories;
f. an adaptive learning mechanism (110) is incorporated within the LNN (108) model, said adaptive learning mechanism (110) being configured to update parameters of the LNN (108) model based on variations in input data; when
g. multispectral satellite data (102) having different spatial and temporal characteristics is received, said adaptive learning mechanism (110) adjusts the LNN model to maintain classification accuracy without requiring complete retraining;
h. the method enables handling of spatial variations across different geographic regions and temporal variations across different time periods to improve robustness and generalization; and
i. wherein said method generates classification output in the form of labeled maps or datasets (112) representing different land use and land cover categories.
2. The method as claimed in claim 1, wherein the preprocessing (104) of the multispectral satellite data (102) comprises noise reduction and correction to improve data quality.

3. The method as claimed in claim 1, wherein the spectral and spatial features (106) include information related to reflectance, texture, and spatial patterns of land regions.

4. The method as claimed in claim 1, wherein said Liquid Neural Network (108) (LNN) model is configured to capture temporal variations present in the multispectral satellite data (102).

5. The method as claimed in claim 1, wherein said adaptive learning mechanism (110) enables continuous updating of parameters of the LNN model without requiring complete retraining of the model.

6. The method as claimed in claim 1, wherein said adaptive learning mechanism (110) is configured to respond to changes in data distribution across different geographic regions.

7. The method as claimed in claim 1, wherein said adaptive learning mechanism (110) is configured to adjust the LNN model based on seasonal or time-based variations in the input data.

8. The method as claimed in claim 1, wherein the classification output includes identification of land categories comprising forest, water, urban, and agricultural land.

9. The method as claimed in claim 1, wherein said method is implemented in a computer system for real-time or near real-time land use and land cover classification.
10. The method as claimed in claim 1, wherein the generated classification output is provided in the form of maps or structured datasets (112) for analysis.

6. DATE AND SIGNATURE
Dated this on 06th April 2026
Signature

Mr. Srinivas Maddipati
(IN/PA 3124)
Agent for applicant