Intellectual Property India Publishes Patent Application for 'Plant Leaf Disease Image Classification Method Integrating Capsule Network And Residual Network' Filed by Ravindra College Of Engineering For Women

MUMBAI, India, Feb. 13 -- Intellectual Property India has published a patent application (202641007503 A) filed by Ravindra College Of Engineering For Women, Kurnool, Andhra Pradesh, on Jan. 26, for 'plant leaf disease image classification method integrating capsule network and residual network.'

Inventor(s) include A. Sai Rekha; S. Saritha; L. Sandhya Rekha; V. Neelima; K. Swathi; and P. Chandana Reddy.

The application for the patent was published on Feb. 13, under issue no. 07/2026.

According to the abstract released by the Intellectual Property India: "Conventional convolutional neural networks (CNNs) often fall short in capturing the spatial orientation and positional relationships of disease lesions on plant leaves, resulting in reduced accuracy and model robustness. To overcome these challenges, this study proposes a novel image classification approach that integrates an optimized residual network (ResNet) with a capsule network (CapsNet). The proposed method enhances the ResNet architecture by replacing its initial convolutional layer with multiple 3 3 filters, thereby improving the extraction of fine-grained features from leaf lesions. Additionally, a channel attention mechanism is embedded within the residual blocks to enable the network to focus on the most relevant features. To maintain spatial integrity, the traditional pooling layer in ResNet is eliminated, helping retain critical positional information. The output from the third residual module is then fed into the capsule network, effectively combining the strengths of both architectures-ResNet's deep feature extraction and CapsNet's spatial awareness and robustness. The resulting hybrid model, referred to as SE-SK-CapResNet, is validated on three widely used datasets: PlantVillage, AI Challenger 2018, and the Tomato Leaf Disease dataset. The model achieves impressive classification accuracies of 98.58%, 95.08%, and 97.19%, respectively. Furthermore, it demonstrates strong performance under image rotations and complex disease patterns, surpassing conventional CNN- based models. These outcomes highlight the proposed method's suitability for accurate and resilient plant disease detection in agricultural settings."

Disclaimer: Curated by HT Syndication.