Development of a lightweight online detection system for impurity content and broken rate in rice for combine harvesters

Abstract

The impurity content and broken rate of rice are important indicators for measuring the performance of a combine harvester and are related to the quality of the operation. On-line detection of impurity content and broken rate during the operation can help the driver to adjust the working parameters of the combine harvester in time, reduce the impurities and crushed grains in the harvest process, improve the harvest quality and ensure food security. However, most of the existing studies can only detect impurities or broken grains under ideal conditions and cannot obtain accurate values for both impurity content and breakage rate, which makes it difficult to be applied to field operation scenarios of combine harvesters. A lightweight detection system has been developed for impurity content and broken rate of rice, it has a simple and reliable structure, and can collect continuous high-quality grain images. The calculation model for impurity content and broken rate was optimized based on the characteristics of grain distribution in the image. The Mask R-CNN was improved to recognize only the number of complete grains without segmentation. Additional edge detection branches were added to improve the segmentation accuracy of the algorithm for impurities and broken grains. The segmentation accuracy of the improved network for broken grains and impurities was increased by 6.13% and 9.19%, respectively. The developed system exhibited a relative error of 7.99% in detecting impurity content during actual field operations, while the error in detecting broken rate was 8.46%. On average, the system was able to obtain a detection result every 1.5 s.