A Modified MobileNetv3 Coupled With Inverted Residual and Channel Attention Mechanisms for Detection of Tomato Leaf Diseases

Abstract

An early detection of tomato leaf diseases is crucial for ensuring high-quality and abundant crop production. There is a growing need for an Android application that can identify these diseases and alert farmers to take timely and effective measures, thereby reducing crop yield losses. While deep learning models have been instrumental in detecting plant leaf diseases, they often involve complex models and significant computational demands to achieve optimal performance. This study introduces a feasible lightweight model optimized for an Android application to detect diseased areas within images. This research focuses on enhancing the efficiency and accuracy of tomato leaf disease detection by modifying mobile-based Convolutional Neural Networks (CNNs). This model employs two parallel network streams based on the core principles of MobileNetv3, utilizing inverted residual blocks (IRBs) to improve accuracy at both low and high-level features, operating across different image dimensions. Additionally, inverted residual connections are incorporated to expand the model’s receptive field. To maximize feature utilization, cross-layer connections are introduced between the two parallel streams, integrating the Efficient Channel Attention (ECA) module to reduce the number of parameters. The model is trained using transfer learning with specific adjustments to minimize detection errors and is fine-tuned using a tomato leaf disease image dataset extracted from Plant Village. This feature integration and data analysis scheme are then deployed to provide a tomato leaf disease detection Android app. Evaluation of the model’s performance is conducted through correct predictions and quality metrics, with the average training accuracy for the ten classes found to be 98.77%. The research aims to assess the model’s effectiveness through comprehensive experiments, ultimately contributing to advancements in agricultural applications.

Topics & Concepts