Title:

ADVANCED DEEP LEARNING METHODS FOR PLANT DISEASE DETECTION

Author:

Rajasree R and Dr. Lilit Kumar Khatri

Abstract:

To reduce financial losses and boost agricultural yield, it is essential to diagnose plant leaf diseases accurately and promptly. However, since farmers still use manual techniques, it might be difficult to accurately diagnose certain illnesses. This study's primary goal was to assess several deep learning techniques for plant pathology picture classification on a large dataset that included 38 distinct classes of conditions related to healthy or sick plants. There are almost 87,000 images in the collection. Therefore, this work presents results employing a specially constructed CNN as well as certain state-of-the-art, pre-trained CNNs such as ResNet50, InceptionResNetV2, and VGG16. According to the findings, VGG16 achieved the highest test accuracy of 96.52%, while InceptionResNetV2 came in second with 94.67%. With a test accuracy of 94.32%, the custom CNN model likewise performed rather well in this study. Regarding several performance criteria, it has also been determined that VGG16 excels at the relevant job. This study examines the possible contributions of CNN fine-tuning and transfer learning to the advancement of agricultural precision.

Keyword:

Deep Learning, Convolutional Neural Networks, Plant Pathology.