Enhanced Convolutional Neural Network (ECNN) for Maize Leaf Diseases Identification

Abstract

Livestock and plants are cultivated by an art and science called agriculture. In sedentary human civilization rise, key development is agriculture. Food surpluses are formed by domesticated species farming. In the absence of automatic diagnosis and identification of maize leaf disease, plants may be collapsed and even tend to die due to leaf disease which affects leaves of a plan to certain value. Vegetable and fruits supply may be drastically decreased by this disease in market. Different detection techniques for plant leaf disease are used in the literatures. Large areas are not covered to detect leaf disease in those methods, and they consumes more time. In maize leaf disease, convolutional neural networks (CNNs) and deep neural networks (DNNs) are used successfully for network parameter reduction and for improving maize leaf disease accuracy of identification. Diagnosis of maize leaf disease is done by enhanced convolutional neural network (ECNN) with receptive field’s enlargement in this research. Four aspects are used to implement ECNN by this research. That includes ECNN framework, fused dilated convolutional layer, convolutional layer with one dimension, and ECNN motivation. Multiple pooling and stacked fused dilated convolutional layers, one input and one-dimensional convolutional layer are composed by ECNN. Estimated and real probability’s cross-entropy is computed at final stage. ECNN weights are updated by a gradient descent method. Epochs of backpropagation are multiplied to compute optimum parameters. Unmodified models are used to make a result comparison of experimentation. Maize leaf disease is identified by proposed method. Google Web sites and plant village are used to gather around 500 images. This collection of images includes maize leaf disease’s various stages. There are 9 classes of those images. Analysis of F-measure, accuracy, recall, and precision parameters is done by experimentation.