The prevention and control of rice leaf diseases require timely and precise detection to enhance crop productivity and minimize the economic damages in precision agriculture. Current cloud-based and deep learning methods are in general characterized by high latency, large computational and low applicability to resource-limited field operational environments. In order to overcome these issues, this paper introduces a proposed solution to the problem called TinyRiceNet-IoT, which is a complete embedded TinyML platform that is able to detect rice leaf disease in real-time. This model uses a depthwise separable convolutional neural network, batch normalization, and global average pooling using a lightweight network that can be run on a microcontroller platform with ease. TinyRiceNet-IoT is an IoT platform combines edge-level intelligence and lightweight IoT communication, which enables transmitting of only small amounts of diagnostic information to the remote environment to generate alarms and monitor. The proposed framework has much lower inference latency, memory footprint and communication overhead than conventional cloud-based CNNs and edge-GPU models, but has competitive classification accuracy. Thorough experimental analysis proves that TinyRiceNet is highly accurate with balanced accuracy, recall, and F1-scores on a variety of rice leaf disease categories and ablation experiments are more successful than simplified model variants. The findings affirm that TinyRiceNet-IoT is a good and scalable solution to real-time monitoring of rice disease in resource constrained farming settings.
P. Sundaravadivel (Mon,) studied this question.