A Comparative Analysis of Deep Learning based Classifiers for the Identification of Maize Rot Diseases

One of the main factors causing production deprivation in specific provinces of India, is infections with maize stalk rot, however large-scale identification of maize illnesses is challenging due to limitations in standard image-recognition technology and a lack of human competency. Thankfully, advances in computer intelligence technology allow for automatic disease identification, which can assist farmers in rapid and dependable disease detection. In the current analysis, we compared the success rate of multiple deep learning-based classifiers for identifying two of the dominant varieties of maize stem diseases in India. In order to do this study, we collected 1364 photos of maize, which fall into three different categories: healthy, Charcoal rot, and Fusarium stalk rot. Furthermore, all these images undergo analysis using 13 distinct pre-trained CNN models including AlexNet, Densenet-201, Googlenet, Inceptionresnet-V2, Inception-V3, Mobilenet-V2, VGG 16 & 19, Resnet 18, 50 & 101, XceptionNet, and ShuffleNet for both training and testing purposes. The performance metrics, namely accuracy and False Positive Rate, are meticulously recorded for each CNN model. Subsequently, the efficacy of every individual model is assessed via multilabel analytical investigation conducted using IBM SPSS tool, aiming to identify the utmost image categorization model among the aforementioned models.