Plant Disease Classification on IoT Edge Devices: Based on modified MobileNetV1 with Transfer Learning Technique

Abstract

An inexorably feasible and a less expensive alternative IoT devices for cloud computing with a ground-breaking approach is Edge Computing. Notwithstanding bringing down the expense of systems administration frameworks, edge processing decreases the delay in edge-cloud, which is fundamental for all crucial applications. Plant diseases disastrously affect the security of the food creation, and that can create a critical decrease both in the quality and in the amount of farming items. In serious circumstances, plant ailments may remain reason for no grain reap totally. Accordingly, all the agriculture data extremely need automatic plant diseases detection and diagnosis system.Here in this paper, we have proposed a transfer learning grounded plant disease detection approach and tried the model's exhibition on IoT edge gadgets. We have utilized MobileNetV1 pre-trained model with ImageNet dataset as base model at that point supplanted the top layer with a convolutional layer and trailed by a softmax classifier. To decrease overfitting, we have included a dropout layer later the convolutional layer. The MobileNetV1 is utilized to separate the features and softmaxtoclassify. The model was trained utilizing Adam optimizingmethod and accomplished 99.77 % classification precision. The dataset comprises of 3 plants and 18 classes of diseases with 41,321 plant leaf images. The presentation of modified MobileNetV1 is additionally contrasted to the other pre-trained models, for example, MobileNetV2 and NASNetMobile. This changed model was executed on Google Colab utilizing TensorFlow and open source transfer learning structure. Since, the MobileNetV1 is a light weight neural network the trained model was sent in low-force and restricted figuring IoT gadgets, for example, raspberry pi and android cell phone.