Plant leaf disease classification based on feature selection and deep neural network

Abstract Today, deep learning (DL) is bringing the big transformation across major industries. Agriculture is one industry where DL scientists and researchers are working with farmers to help farmers make a better and more efficient use of the shrinking resources due to urbanization. However, plant disease, especially to crop plants, is a major threat to the global food security. Plant diseases directly affect the quality of the fruits, grains, and so on, leading to the decrease of agricultural productivity. The traditional method of identifying plant disease is done by visual examination. This process is highly inefficient and error prone. In recent years, several works on DL techniques for leaf disease have been proposed. Most built their models based on limited resolution images on convolutional neural networks. In this chapter, we want to focus on early disease recognition, which requires higher-resolution images. After a preprocessing step using a contrast enhancement method, all the diseased blobs are segmented for the whole data set. A list of several measurement-based features that represent the blobs is chosen and selected based on principle component analysis. The features are used as inputs for a standard feed-forward neural network. Our results not only show competitive classification results with other DL approaches such as convolutional neural networks, but also with a simpler network structure.