Evolutionary Approach-Based Hybrid CNN and SVM for Effective Plant Disease Classification

Agriculture is one of the prime factors that form the backbone of the economy of India. Also, with its rapidly growing population, the demand for food is increasing manifold. Thus, there arises a dire need to augment agricultural productivity to meet this ever-growing demand. One of the highest challenges that prevents quite ideal crop yields from being acquired is the prevalence of plant diseases, which could be due to many kinds and types of pathogens. For example, bacteria, viruses, fungi, and others are at fault. Thus, the discovery of effective detection systems becomes essential to combat their ill effects. One of the most promising strategies for plant disease identification is using machine learning algorithms. These techniques have proved to be effective in analysis resulting from big data, mainly due to their focus on reaching specific specified results that come out perfectly for identifying plant diseases. Worth noting is that the successful application of machine learning in this setting often requires feature engineering. This is a process that involves the choice and modification of data features likely to enhance the performance of the model. In this paper, we would like to propose a different method of plant disease recognition using convolutional neural networks. The modification of the last layer of the conventional Softmax CNN model opens up an opportunity for enhancing the precision of the approach, done by substituting the usually used Softmax classifier with one based on the k-nearest neighbors algorithm. This will lead us toward better results. The modification uses the ability of this K-NN classifier to work along the labels of the neighboring data points, thus giving better accuracy in decision-making concerning disease detection. Moreover, we have developed an evolutionary algorithm to adjust the parameters of the network for the improvement of accuracy in CNN's performance. This systematic searching technique is one kind of evolutionary technique that seeks to find the appropriate configuration of CNN parameters, which maximizes the model efficiency on plant disease identification tasks.