Detection of Plant Diseases Using Convolutional Neural Network Architectures

Abstract

: Plant diseases will wreak havoc on agricultural products' quality and quantity. It is important to recognize plant pathogens early on for the sake of global health and well-being. Deep learning's popularity in machine vision has recently inspired many researchers to improve the performance of plant disease detection systems. Unfortunately, most of these studies relied on AlexNet, GoogleNet, and other similar structural design rather than more recent deep designs. Furthermore, the research did not employ deep learning visualization techniques, which classify deep classifiers as "black boxes" due to their opacity. We used these three learning techniques to assess various state-of-the-art Convolutional Neural Network (CNN), AlexNet, and VGG16 architectures on a public dataset for plant disease classification in this article. In comparison to other designs, the VGG16 outperforms state-of-the-art findings in plant disease classification, with an accuracy of 97 percent. In addition, we have suggested the use of saliency maps as a means of visualizing and interpreting the CNN classification mechanism. This method of visualization improves the clarity of deep learning models and provides further insight into plant disease symptoms. This paper compares the disease classification of CNN, AlexNet, and VGG16 designs on mangoes, grapes, potatoes, rice, and corn leaves. In comparison to CNN and AlexNet designs, the VGG16 architecture has high accuracy and recall. Precision separates predictive positive from actual positive, while recall separates actual positive from predictive, positive, and high precision and recall mean that the classifier is generating accurate results. The next project for the research team will be to develop a smartphone app that will diagnose the disease and be useful to farmers. Farmers will photograph diseased leaves, and the mobile device will identify the issue and include instructions about how to address it. It would be very good for farmers with vast fields because it will be more efficient and less time intensive. DL designs have also been discovered to be capable of identifying essential and irrelevant features from a series of images through this research.