Automation of Grain Quality Assessment Processes Using Technical Vision and Neural Networks in Mechanical Engineering and the AIC

This study explores modern methods for automated grain quality assessment using computer vision and machine learning technologies. The primary focus is on the problem of detecting weed impurities and broken grains in harvested crops, which is critical for improving the efficiency of agricultural production. A comparative analysis of various approaches is conducted, including classical algorithms (RANSAC, OpenCV) and modern neural network architectures (YOLOv11x, U-Net, ResNet18). The research results demonstrate that neural network methods, particularly YOLOv11x, achieve the highest accuracy (up to 95%) in detecting grain impurities and breakage, outperforming traditional computer vision algorithms. Of particular importance is the potential integration of the developed solutions into combine harvester systems for real-time grain quality monitoring during harvesting and yield mapping. A hybrid approach is proposed, combining image preprocessing with OpenCV and subsequent detection using neural network models.