A hybrid framework for detection of diseases in apple and tomato crops with deep feed forward neural network

Abstract

The traditional farming methods with low productivity and crop damage due to diseases have resulted in low economic growth of the farmer. To overcome this problem, an expert system capable of monitoring the crop growth and early detection of the diseases is highly essential for the farmer to take preventive steps. In this work, firstly, the quality of the image is improved using both threshold-based and principle component analysis techniques. Then the enhanced images were divided into three individual colour channels (red, green, and blue) and then 27 statistical features comprising of texture, colour and energy are computed. The same statistical features of the first-level wavelet decomposition are appended to those 81 features. Finally, the features were classified by the deep feed forward neural network to identify type of the disease. The proposed method outperformed the existing methods by yielding 94.06% and 96.78% of accuracy on tomato and apple datasets respectively.