A CBAM-MobileNext-based Method for Plant Disease Identification

Focusing on the problems of slow response speed, large number of parameters, and large consumption of computer memory resources in the existing plant disease recognition models, this study proposes a lightweight neural network model, CBAM-MobileNext, which selects MobileNext as the backbone network, and introduces the attention mechanism of CBAM and the Ghost module to improve the original model's SandGlass module. To validate the positive aspects of this model, its performance is evaluated alongside non-lightweighted deep learning models (AlexNet, VggNet-16) and lightweight models (GoogLeNet, MobileNet V2) on the PlantVillage dataset, and the final inference results of the model show that,CBAM-MobileNext in terms of floating-point operations, accuracy, single image inference time, number of parameters, F1 value and model size, which can accurately recognize plant diseases in a relatively brief span of time and can be deployed on mobile devices with limited resources, providing technical support for intelligent disease recognition in real agricultural scenarios.