Implementation of Leaf Disease Detection Using One-Shot & Region Inception Image Recognition Technique

The rate of economic expansion is directly related to agricultural production. The presence of illness in plants is quite widespread, which is one of the reasons why plant disease detection is important in the agricultural industry. When safeguards are not taken in this area, plants incur severe effects that affect the quality, quantity, or productivity of the associated products. In order to monitor big crop farms with minimal manpower and to detect disease signs as soon as they first appear on plant leaves, it is preferable to employ an automated technique for plant disease detection. Visual features play an important role to find diseases in plants through leaves. Visual features along with deep learning techniques are a growing area of research and application in today's era. Industries like Facebook AI research contributed to deep learning and self-learning model in the last few years. This paper presents a work carried out on a self-learning model to detect and classify defective plants using CNN with a Siamese network. Large datasets of annotated data are often needed for convolutional neural networks but are rarely available on demand. It takes a lot of time and effort to personally choose, photograph, and annotate each leaf to get this data. This work addresses the issue of limited plant picture data by examining the effectiveness of various data augmentation methods when combined with transfer learning. This paper systematically showcases results related to visual feature representation, similarity computation, and experimental to compare the proposed work. Our goal with paper is to bridge the performance gap with lot many existing techniques of deep learning.