Abstract
In this article, a soybean process optimisation solution using real-time artificial intelligence of things (RT-AIoT) technology at the edge is presented. Image classification, object detection and recognition are machine vision techniques implemented into industrial internet of things (IIoT) devices to determine variations in the morphological features in soybeans. Evaluating soybean features, such as moisture and temperature combined with other measurements, such as colour, size, shape, and texture, can improve the utilisation of the raw material and the quality of the derived products, thus reducing energy consumption. Implementing intelligent vision locally on IIoT edge devices solves several issues faced by deploying it to the cloud and brings further challenges posed by deep learning on resource-constrained edge devices. Most deep neural networks are too complex to be created and trained on most nowadays microcontrollers, but if optimised in terms of memory, processing, and power capabilities, they can run on them. With multi-image sensors, and IIoT devices under evaluation, the proposed production optimisation system is interfaced with the existing industrial SCADA system, and analyses the IIoT sensor data at different edge computing granularity levels. With the preliminary findings and results, we show that the RT-AIoT, including machine vision technology, is now possible on all micro, deep and meta edge levels with the advent of AI.