An Innovative Onion Disease Classification Framework Using Smart IoT Assisted Multi-Scale Attention-Based ResNeXt with SPP via Modified Cat–Mouse Optimizer

The application of smart agriculture is enabled by the Internet of Things (IoT) and cloud-based platforms, and these techniques have reshaped conventional methods of farming. Producers may prevent onions from spoilage by using storage procedures. However, humans can fail to identify the decaying onions, in which case managers may depend on sensors and deep learning strategies for detecting the decomposition of onions. The conventional techniques estimate data improperly because of external factors, and the system that covers the IoT framework encountered several challenges while using the currently accessible techniques to recognize onion disease. Hence, an IoT-based onion disease classification framework using deep learning is proposed for increasing the onion yield by recognizing the onion disease. At first, the required data are collected using the IoT devices. Further, the region of interest (ROI) detection process is done using the developed adaptive residual fully convolutional network (ARFCN), and finally, the ROI cropped images are attained with the support of the ARFCN technique. Here, the variables of the ARFCN approach are tuned with the help of a revised random parameter-based cat and mouse optimizer (RRP-CMO). The preservation of valuable regions is highly augmented by the ROI detection process. In addition to that, the acquired image from the dataset has few unwanted regions, and it is removed by the ROI detection process. The cropped images are then preceded for disease classification using the proposed multi-scale attention-based ResNeXt with spatial pyramid pooling (MARSPP). The final classification results are validated with various conventional algorithms and classifiers.