Detection of disease in tomato plant using Deep Learning Techniques

Abstract

With continual advances in technology, there is an increased demand to provide enough food for more than seven billion people. This work is limited to automated detection of disease in cultivated land as it is a serious threat to food production and affects the livelihood of the small-scale farmers. In conventional farming practices, skilled people are employed to scout the land manually and detect the presence of disease in the land by visual inspection. The row by row manual detection of diseases in cultivated land is laborious and time-consuming. At times, the laborious work is prone to error. With the help of advanced image processing techniques and algorithms, this work aims at developing an automated mechanism for detecting the diseases in cultivated land. Deep learning techniques, specifically with deep detector: Faster R-CNN with deep feature extractor: ResNet50, is used to detect and classify tomato disease in plants. Trained and tested the proposed system with our tomato dataset, which has 1090 comprehensive images of early, medium, and final stages of tomato disease. Our proposed system successfully detects early blight, leaf curl, septoria leafspot, and bacterial spot of tomato disease even in complex plant surrounding areas.