Published Paper Details:

The early prevention of crop diseases and identifyingthem at early stage is very important for crop production. The usage of fertilizers and chemicals is causing a negative effect on the agriculture eco-system and causes economic loss. A reliable method to identify and diagnose the diseases will benefit the farmers. In this paper, a new model is introduced with depthwise separable convolution architecture for tomato plant disease detection using leaf images. The models were trained and tested on a subset of publicly available plant village dataste containing six different classes of healthy and diseased leaves. These depthwise separable convolutions achieved high gain in convergence speed and the accuracy is also improved. A novel model called CNNLDD is proposed and it is compared with reduced CNNLDD model and this reduced model achieved a classification accuracy of 98.39% with fewer parameters. The satisfactory accuracy and small size of this model makes it suitable for real time crop diagnosis with less computation cost and the training time is also minimized. The standard CNN model requires a huge number of parameters, computation cost is high and also the training time is more. Standard convolution is replaced with the depth wise separable convolution in order to minimize the number of parameters and computation cost.