DEEP LEARNING MODELS FOR CROP DISEASE DETECTION USING IOT-ENABLED IMAGING SYSTEMS

Abstract

This paper addresses the growing challenge of accurate and timely crop disease detection of rapid and proper detection of crop diseases in large scale farming where manual observations are inefficient, subjective and expensive. The aim is to have an IoT enabled imaging system coupled with deep learning models to perform automated, scalable and real time diagnosis of crop diseases. The proposed system captures crop images using field-deployed IoT cameras, transmitting data up to clouds and smart classification of diseases in various crops. The proposed solution involves the combination of convolutional neural networks, attention, transfer learning, and data augmentation in order to increase resilience to changing illumination and background noise. It compares architectures of ResNet, EfficientNet, and attention embedded CNN, and GAN based image augmentation and optimization contribute to the enhancement of generalization. Experimental results demonstrate that the proposed framework achieve that the new framework can provide classification, accuracy of 96.8, precision of 95.9, recall of 96.1, and F1 score of 96.0, beating traditional CNN models by 7-12 percent in the main metrics. Edge assisted inference is able to decrease latency by 34% as compared to cloud only deployment which makes it possible to respond in near real time. The results affirm that IoT imaging with optimized deep learning model has a high potential in enhancing the reliability of detection, scaling, and response time. The paper concludes that the suggested IoT based deep learning system is a successful solution to precision agriculture to assist in early disease detection, less crop loss, and sustainable smart farming methodology with high deployment opportunities in the resource limited agricultural settings of the world.