Tomato Leaf Diseases Detection with Recommended Prescription Using Deep Learning

Detecting early signs of plant leaf diseases is vital in an agrarian economy. Automatic Leaf disease recognition is extremely important because predictive mechanisms that aid in the avoidance of losses can be adopted timeously. Transfer learning algorithms have become a popular solution for recognizing tomato leaf diseases over the years. However, they have innate limitations which include higher processing time and lower accuracy according to various expectations in regards to problem domain. A classification algorithm is as good as the features used to develop the model. Therefore, the primary objective of this experimental study was to discover the most discriminating features in detecting tomato leaf diseases. The paper presents a combination of Grayscale Pixel Value (GPV) features and ResNet9 in an effort to solve the aforementioned problem. We evaluated the proposed solution against other features such as Mean Pixel Value (MPV) and other deep learning generated features. The results showed that our proposed method is effective in detecting tomato leaf diseases because of the significantly low computation time (10 minutes) and superior accuracy (98.59%).