Description:FIELD OF THE INVENTION
This invention relates to A Region Proposal Network with Adaptive Scale Non- Maximum Suppression methods in Deep Learning Model for Tomato Leaf Disease Detection
BACKGROUND OF THE INVENTION
Timely prediction of tomato plant diseases is essential to reduce crop losses, supporting better harvest and increasing the productivity. Since, Deep Learning models play an important role in plant diseases prediction, but face certain challenges. The images taken under various lighting conditions like daylight, dawn, dusk and evening highly affects the image intensity. This makes difficult for the deep learning models to learn features like small spot Region of Interest (ROI), color, gradients, textures and shape from disease regions. These variations between training and testing image scan create misleading results and in some cases, important details are lost due to low contrast backgrounds. All these challenges significantly lower the models efficiency in tomato plant diseases prediction. The existing FRCNN used for tomato plant diseases prediction has issues of redundant region proposals generated by Region Proposal Network (RPN)
SUMMARY OF THE INVENTION
This summary is provided to introduce a selection of concepts, in a simplified format, that are further described in the detailed description of the invention.
This summary is neither intended to identify key or essential inventive concepts of the invention and nor is it intended for determining the scope of the invention.
To further clarify advantages and features of the present invention, a more particular description of the invention will be rendered by reference to specific embodiments thereof, which is illustrated in the appended drawings. It is appreciated that these drawings depict only typical embodiments of the invention and are therefore not to be considered limiting of its scope. The invention will be described and explained with additional specificity and detail with the accompanying drawings.
To solve the problems of tomato plant diseases from tomato leaves, inspired by FRCNN, a novel network structure is proposed with a twofold pathway hierarchy structure and ROI refinement model. This twofold pathway hierarchy combines top-down feature extraction and bottom-up attention to capture both high-level semantics and fine-grained features. Feature Pyramid Network (FPN) is used to extract multi-scale features (top-down) while spatial pyramidal attentions (bottom-up) highlight important regions. In order to find potential relevant disease regions in bottom-up attention mechanism, a novel Region Proposal Network (RPN) is prosed to generate discriminative ROIs with compatible sizes. An Adaptive Scale Non- Maximum Suppression (ASNMS) is applied on RPN for reducing redundancy and maintaining visual integrity of the images. Classical NMS relies on a fixed Intersection over Union (IoU) threshold, which is inefficient for variable lesion sizes and shapes throughout disease-affected areas. Smallordispersed lesions often have insufficient IoU, resulting in missed suppression and false positives. ASNMS accomplishes this by scaling predicted bounding boxes depending on the particular disease category, guaranteeing that even tiny or irregularly shaped diseases regions provide sufficient overlap for appropriate suppression and thereby improving detection accuracy. Thus, spatial pyramidal attentions masks generate ROIs effectively in discriminative parts of the images. Moreover, ROI refinement uses two operations as if ROI drop block removes the most discriminative small-scale regions to prevent overfitting, while ROI merge block combines all ROIs to highlight the major regions for better feature representation. These refined features are then passed through the Softmax classifier for classification. This approach enables the model to learn robustly even under poor lighting conditions to improve the prediction accuracy, handles small spot detection and enhances reliability in real-world agricultural settings.
BRIEF DESCRIPTION OF THE DRAWINGS
The illustrated embodiments of the subject matter will be understood by reference to the drawings, wherein like parts are designated by like numerals throughout. The following description is intended only by way of example, and simply illustrates certain selected embodiments of devices, systems, and methods that are consistent with the subject matter as claimed herein, wherein:
FIGURE 1: SYSTEM ARCHITECTURE
The figures depict embodiments of the present subject matter for the purposes of illustration only. A person skilled in the art will easily recognize from the following description that alternative embodiments of the structures and methods illustrated herein may be employed without departing from the principles of the disclosure described herein.
DETAILED DESCRIPTION OF THE INVENTION
The detailed description of various exemplary embodiments of the disclosure is described herein with reference to the accompanying drawings. It should be noted that the embodiments are described herein in such details as to clearly communicate the disclosure. However, the amount of details provided herein is not intended to limit the anticipated variations of embodiments; on the contrary, the intention is to cover all modifications, equivalents, and alternatives falling within the scope of the present disclosure as defined by the appended claims.
It is also to be understood that various arrangements may be devised that, although not explicitly described or shown herein, embody the principles of the present disclosure. Moreover, all statements herein reciting principles, aspects, and embodiments of the present disclosure, as well as specific examples, are intended to encompass equivalents thereof.
The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments. As used herein, the singular forms “a",” “an” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms “comprises,” “comprising,” “includes” and/or “including,” when used herein, specify the presence of stated features, integers, steps, operations, elements and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and/or groups thereof.
It should also be noted that in some alternative implementations, the functions/acts noted may occur out of the order noted in the figures. For example, two figures shown in succession may, in fact, be executed concurrently or may sometimes be executed in the reverse order, depending upon the functionality/acts involved.
In addition, the descriptions of "first", "second", “third”, and the like in the present invention are used for the purpose of description only, and are not to be construed as indicating or implying their relative importance or implicitly indicating the number of technical features indicated. Thus, features defining "first" and "second" may include at least one of the features, either explicitly or implicitly.
Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which example embodiments belong. It will be further understood that terms, e.g., those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.
To solve the problems of tomato plant diseases from tomato leaves, inspired by FRCNN, a novel network structure is proposed with a twofold pathway hierarchy structure and ROI refinement model. This twofold pathway hierarchy combines top-down feature extraction and bottom-up attention to capture both high-level semantics and fine-grained features. Feature Pyramid Network (FPN) is used to extract multi-scale features (top-down) while spatial pyramidal attentions (bottom-up) highlight important regions. In order to find potential relevant disease regions in bottom-up attention mechanism, a novel Region Proposal Network (RPN) is prosed to generate discriminative ROIs with compatible sizes. An Adaptive Scale Non- Maximum Suppression (ASNMS) is applied on RPN for reducing redundancy and maintaining visual integrity of the images. Classical NMS relies on a fixed Intersection over Union (IoU) threshold, which is inefficient for variable lesion sizes and shapes throughout disease-affected areas. Smallordispersed lesions often have insufficient IoU, resulting in missed suppression and false positives. ASNMS accomplishes this by scaling predicted bounding boxes depending on the particular disease category, guaranteeing that even tiny or irregularly shaped diseases regions provide sufficient overlap for appropriate suppression and thereby improving detection accuracy. Thus, spatial pyramidal attentions masks generate ROIs effectively in discriminative parts of the images. Moreover, ROI refinement uses two operations as if ROI drop block removes the most discriminative small-scale regions to prevent overfitting, while ROI merge block combines all ROIs to highlight the major regions for better feature representation. These refined features are then passed through the Softmax classifier for classification. This approach enables the model to learn robustly even under poor lighting conditions to improve the prediction accuracy, handles small spot detection and enhances reliability in real-world agricultural settings.
Most existing deep learning models fails to address issues like not focusing on small detailed lesions in leaf surfaces or variation in image intensity owing to lighting, as there is a lack of consistent illumination and focus on small detailed regions is difficult. Such models do not perform well in practical agricultural settings since they rely on high-quality, paired, and large-volume datasets. In contrast, the proposed model introduces a twofold pathway hierarchy structure combined with an ROI refinement model that enhances both low-level feature extraction and accurate localization of disease regions, even under poor lighting conditions. Earlier models failed to focus on micro features and small ROIs, so this model uses advanced FPN model to extract features across multiple scales, novel RPN with ASNMS on spatial pyramidal attention to address these shortcomings. Additionally, the integration of ROIs drop block and merge operations allow to extract contextual relevant disease regions that boost the models performance. These integrations not only reduce the need for extensive labeled datasets, therefore increasing scalability and economically benefiting precision agriculture.
NOVELTY:
The traditional models struggle with poor image intensity, low-resolution features and inaccurate localization of disease-affected areas, lowering the models performances. To solve this, twofold path way hierarchy structure combined with an ROI refinement strategy. The dual- pathway hierarchy uses a top-down feature pathway and a bottom-up attention pathway to enhancelow-levelfeaturerepresentationandimprovesemanticinformationflow. FPN model to extract features across multiple scales. The model uses a novel RPN with ASNMS based on spatial pyramidal attention masks captures fine-grained disease features. ASNMS effectively scales the predicted boxes based on disease type, allowing small or irregular lesions to generate enough overlap for effective performance. Additionally, ROI drop block and ROI merge block operations refine the regions by eliminating irrelevant noise and focusing on meaningful disease patterns. The novelty highlighted in this work assist to eliminate and solve the intensity problem due to different lighting conditions and handling small ROI of disease region effectively localize various discriminative regions for accurate tomato plant diseases prediction.
 
, Claims:1. A method for detecting tomato leaf diseases using a deep learning model,
comprising:
a twofold pathway hierarchy structure integrating a top-down feature extraction pathway and a bottom-up attention mechanism;
wherein the top-down pathway utilizes a Feature Pyramid Network (FPN) for extracting multi-scale semantic features;
and the bottom-up pathway uses spatial pyramidal attention masks to highlight fine-grained regions of interest (ROIs);
characterized in that a novel Region Proposal Network (RPN) is used to generate discriminative ROIs of compatible sizes, and
an Adaptive Scale Non-Maximum Suppression (ASNMS) is applied to said RPN outputs to dynamically suppress redundant ROIs based on disease type, thereby enhancing localization and reducing false detections.
2. The method as claimed in claim 1, wherein the ASNMS scales predicted bounding boxes based on the category of tomato disease,
such that even small or irregularly shaped lesion regions provide sufficient overlap for effective suppression,
thereby improving detection accuracy across variable lesion sizes and shapes.
3. The method as claimed in claim 1, wherein the ROI refinement model comprises:
a ROI drop block configured to remove highly discriminative small-scale regions to prevent model overfitting;
and a ROI merge block configured to combine all identified ROIs to enhance major disease region representations,
whereby improving contextual understanding and feature robustness under varying image lighting conditions.
4. The method as claimed in claim 1, wherein the spatial pyramidal attention masks are generated in the bottom-up pathway to guide the RPN in proposing disease-specific ROIs,
thereby improving the model’s sensitivity towards micro features and enhancing detection performance in low-resolution or poorly illuminated images.
5. A computer-implemented system for tomato leaf disease detection,
comprising a deep neural network configured with:
(a) a Feature Pyramid Network (FPN) for multi-scale feature extraction;
(b) spatial pyramidal attention modules for bottom-up focus;
(c) a Region Proposal Network (RPN) adapted to generate size-compatible ROIs;
(d) an Adaptive Scale Non-Maximum Suppression (ASNMS) module applied to said RPN outputs for dynamic redundancy removal;
(e) an ROI refinement unit including a drop block and merge block; and
(f) a classification layer with Softmax activation for final disease classification;
wherein the system is adapted to operate robustly under inconsistent lighting and enables effective localization of small and irregular disease features.