基于多阈值图像分割算法的秸秆覆盖率检测

针对目前秸秆覆盖率人工检测费时费力、准确率低、信息难以存储的问题，提出了一种基于图像分割的秸秆覆盖率检测方法。考虑到传统图像分割方法精度不高，且多阈值分割时计算量过大，将灰狼算法中的搜索机制与差分进化算法相融合，提出一种基于图像多阈值的自动分割方法（DE-GWO），用于田间秸秆覆盖率检测。首先，对现场采集的秸秆覆盖图像进行预处理，采用自适应Tsallis熵作为目标函数，评估图像分割效率；其次，根据图像的复杂程度选取分割阈值的数量，利用DE-GWO算法对其进行多阈值图像分割；然后，分别按照灰度级别计算分割后图像比例；最后，根据拍摄高度、fov视角等参数，将图像中秸秆覆盖率与实际地理面积进行转换。实验结果表明，本文算法田间秸秆覆盖率与实际测量误差在8%以内，且相比于改进粒子群算法（PSO）和灰狼算法（GWO），DE-GWO算法精确度更高，平均耗时为人工测量的1/1500。开发了一套依据DE-GWO算法的秸秆覆盖率检测软件系统，为后续监控系统的实时检测提供了算法基础和软件支持。

Detection of Straw Coverage Rate Based on Multi-threshold Image Segmentation Algorithm

Straw returning is one of the most important measures for increasing fertility. But straw returning has not been widely popularized at present. It needs to be supervised and tested. However, manual detection of straw coverage is time-consuming, laborious, low accuracy and difficult to store information. In order to solve these problems, a straw coverage detection method was proposed based on image segmentation. Considering the precision of traditional image segmentation method was not high, and the computation was complex for multi-threshold segmentation, the search mechanism of gray wolf (GWO) algorithm and differential evolution (DE) algorithm were combined, and a multi-threshold automatic segmentation method was proposed based on image, DE-GWO algorithm for field straw mulching detection. Firstly, the straw mulching image collected in the field was preprocessed, and the adaptive Tsallis entropy was used as the objective function of the algorithm to evaluate the efficiency of image segmentation. Secondly, the number of segmentation thresholds was selected according to the complexity of the image, and the multi-threshold image was segmented by DE-GWO algorithm. The proportion of the images after the segmentation was calculated by the gray degree level. Finally, the straw mulching rate in the image and the actual geographic area were converted according to the shooting height and the wide angle of the camera. The experimental results showed that the straw mulching rate in the field and the actual measurement error were less than 8%, and the DE-GWO algorithm was more accurate than the improved particle swarm optimization (PSO) and gray wolf algorithm (GWO). Compared with manual measurement, the average consumption time was reduced by more than 1500 times. In addition, a set of software system for detection of straw coverage based on DE-GWO algorithm was developed, which provided the basis of algorithm and software support for the real-time detection of the monitoring system.