ProThinNet23: A Hybrid Deep Learning Model to Detect Diseases in Potato, Tomato and Pepper

Abstract

Potato, tomato, and pepper are the solanaceous crops widely used across the world. These crops are rich source of vitamins, minerals, and fibers. However, many losses occur due to different diseases found in these crops. It is the need of hour to detect these diseases at early stages so that crop losses can be reduced. Conventional methods such as classification models of machine learning do not extract features from its own. On the other hand, deep learning methods are costly in terms of implementation as they require high end computational resources. In the proposed work, integration of both the techniques (machine learning and deep learning) is done. The architecture provided in this article is for a modified version of the ResNet-like neural network in conjunction with optimized Random Forest. The model is trained and tested on Plant Village Dataset consisting of healthy and diseased images of potato, tomato and pepper.  This network aims to provide a lightweight yet effective convolutional neural network (CNN) architecture for the classification of images into different categories. The novelty of proposed hybrid model ‘ProThinNet23’ lies in its ability to provide effective feature extraction and classification while maintaining a relatively small number of parameters. This makes it suitable for scenarios where computational resources are limited, such as on edge devices or mobile applications. The proposed model strikes a balance between model size and performance, providing an alternative to more heavyweight architectures like ResNet50 or VGG16.