Comparative Analysis of Machine Learning Techniques for Plant Disease Detection-Data Deployment

• Published in: Journal of the Institution of Engineers (India). Series B, Electrical Engineering, Electronics and telecommunication engineering, Computer engineering Electrical Engineering, Electronics and telecommunication engineering, Computer engineering, 2023, Vol.104 (4), p.837-849
• Main Author: Deepti, Kakarla
• Format: Article
• Language: English
• Subjects: Accuracy; Communications Engineering; Computer vision; Deep learning; Economic impact; Engineering; Image processing; Impact analysis; Insects; Machine learning; Medical imaging; Networks; New technology; Original Contribution; Plant diseases; Signs and symptoms; Training
• ISSN: 2250-2106; 2250-2114
• DOI: 10.1007/s40031-023-00897-w

Plant diseases and pernicious insects are a huge danger to food security and agriculture sector. It is crucial to identify and recognize the type of plant disease to help the farmer. This information can help to make appropriate decision about an increase in the crop productivity. The detection process marks the beginning of a series of activities to fight the diseases and reduce their spread. The technology used in medical procedures has not been adequate to detect all diseases on time, and that is why some diseases turn out to become pandemics because they are hard to detect on time. To address these issues, new technologies that use image processing, computer vision and deep learning approaches to identify various illnesses in plants are being developed. The results of these methods have shown that they can produce fast, accurate disease detection with a good economic impact. In the present work, a comparison of the performances of various CNN architectures—AlexNet, LeNet, VGG 16 and a novel proposed architecture based on evaluation metrics over two datasets is made out. Analyzing the performance of various models, it is concluded that the VGG 16 is the best architecture with an accuracy of approximately 96%. AlexNet architecture resulted an accuracy of 95%, but the training time was comparatively large. The proposed architecture is a modified LeNet model which has a comparatively less accuracy than VGG 16 and AlexNet but the training time is very less and is reduced by 90% compared to AlexNet. The present work also included creating a user interface to upload the diseased leaf image and get the required diagnosis report including the disease name, its causes, symptoms and treatment suggestions.