Classification of capsicum leaf disease from a complex cluster of leaves using an improved multiple layers ShuffleNet CNN model

Abstract

Capsicum is broadly cultivated throughout the world and is economically substantial as a condiment, vegetable, and medicine. One of the profoundest subjects confronting the capsicum cultivation is the requisite for accurate identification of leaf diseases. Leaf diseases have an adverse effect to the quality of capsicum production, resulting in significant losses for farmers. There has been numerous machine learning (ML) algorithms and convolution neural network (CNN) models developed for classifying capsicum leaf diseases under uniform background with uncomplex leaves conditions, with an average accuracy of classification. However, a diseased leaf typically grows concurrently with a cluster of other leaves and it is relatively challenging to classify the disease. If there is a reliable model that can classify a capsicum leaf disease in a cluster of leaves, of course it will make it easier for farmers. Thus, the aim of this study was to propose a model to classify capsicum leaf disease from a uniform background as well as from a complex cluster of leaves. Firstly, the dataset images of diseased capsicum leaf are acquired, which comprises discolour leaf, grey spot, and leaf curling. An improved multiple layers ShuffleNet cnn model is then utilized to classify the different types of capsicum leaf disease. The proposed model successfully outperformed other existing models, with a classification accuracy of 99.30%. It is also concluded that adding layers to ShuffleNet, a 0.01 initial learn rate, 50 maximum epochs, 64 minibatch sizes, 10 iterations, and 205 validation iterations all contribute to an improved ShuffleNet model.