用多阈值多目标无人机图像分割优化算法检测秸秆覆盖率

为了适应航拍采集秸秆覆盖图像大尺度处理需求，提高当前多阈值差分灰狼优化算法（Differential Evolution Grey Wolf Optimizer，DE-GWO）的图像分割质量和速度，提出一种用于检测秸秆覆盖率的图像分割优化算法。该研究借鉴了人工蜂群多目标灰狼优化算法（Artificial Bee Colony Survey Multi-Objective Grey Wolf Optimizer，AS-MOGWO），在DE-GWO算法中加入了多目标灰狼优化算法（Multi-Objective Grey Wolf Optimizer，MOGWO）的外部存档，引入多目标的概念，并添加了人工蜂群算法（Artificial Bee Colony，ABC）中观察蜂的搜索策略，提出了基于多阈值的多目标秸秆覆盖图像自动分割的优化算法（Differential Evolution Artificial Bee Colony Survey Multi-Objective Grey Wolf Optimization，DE-AS-MOGWO）。该算法不仅继承了DE-GWO算法的自动分割特性，还兼备AS-MOGWO算法的高效收敛性，提高了图像分割的准确性和处理速度。分析结果显示，在无外界影响的情况下，该研究提出的DE-AS-MOGWO优化算法与人工实际测量法匹配的误差可控制在8%以内。在算法性能方面，DE-AS-MOGWO相比于PSO（Particle Swarm Optimization）、GWO（Grey Wolf Optimizer）、DE-GWO和DE-MOGWO在平均匹配率上分别提高了4.967%、3.617%、2.188%和3.404%，平均误分率分别降低了0.168%、0.131%、0.089%和0.116%，而算法耗时分别降低了82%、84%、17%和32%。试验结果表明，多阈值多目标图像分割方法在大尺度无人机图像中可获得较好的分割效果，且针对不同秸秆覆盖率图像均具有普遍适用性，为大面积秸秆覆盖率检测以及其他相关图像检测提供了高效算法支持。

Detection of straw coverage based on multi-threshold and multi-target UAV image segmentation optimization algorithm

Abstract:Straw mulching has been an efficient solution to reduce soil loss in environmental protection and sustainable development in modern agriculture. Therefore, a rapid detection of straw coverage can contribute to the efficiency and accuracy in the process of straw mulching. In this study, a novel algorithm was proposed to optimize large-scale image segmentation for the aerial image of straw coverage during straw mulching. An artificial bee colony survey multi-objective grey wolf optimization algorithm (AS-MOGWO) was used to upgrade via introducing the design concept of multi-objective integration. Specifically, an external archive of multi-objective grey wolf optimization algorithm (MOGWO) was added into the differential evolution (DE) GWO, and the search strategy of observed bees in artificial colony algorithm. The DE algorithm can be used to solve the problem that the traditional gray wolf optimization algorithm (GWO) is easy to fall into the local optimal and the slow processing speed. Extending to multi-objective can also improve the accuracy of multi-threshold image segmentation. The Observation phase of artificial bee colony algorithm (ABC) can be used to compare the solution of problem, and further to enhance the stability and optimization ability of algorithm. The DE-GWO algorithm was extended from single target to multi-target DE-MOGWO, thereby to achieve multi-objective optimization. The accuracy of multi-threshold image segmentation was greatly improved, while, the algorithm was enhanced to extract and classify different ground objects in the collected images. The observation phase of ABC algorithm was added in the detection of straw coverage, further to improve the quality and processing speed of automatic image segmentation. The stability and optimization ability of algorithm can be enhanced after the integration of various methods. The upgraded algorithm inherited the automatic segmentation of DE-GWO, while gained the efficient convergence of AS-MOGWO, indicating an improved stability and processing speed for image segmentation. An optimal threshold was set using the gray-scale histogram of straw image, then to segment the images, and finally to calculate the number of pixels in each part and the coverage of straw. The experimental results showed that the matching error was less than 8% between the DE-AS-MOGWO optimization algorithm and the manual measurement method. Compared with the PSO, GWO, DE-GWO, and DE-MOGWO algorithms, the average matching rate of DE-AS-MOGWO improved 4.967%, 3.617%, 2.188%and 3.404%, respectively, whereas, the average error rate reduced 0.168%, 0.131%, 0.089%and 0.116%, respectively. Furthermore, the algorithm time reduced 82%, 84%, 17% and 32%, respectively. A software system was also developed for the area detection of straw coverage based on the proposed algorithm, where the straw covering area and straw coverage rate can be calculated from the acquisition area of aerial images. The GWO, DE-GWO, DE-MOGWO and DE-AS-MOGWO algorithms can also be selected for the comparison of different results. The DE-AS-MOGWO algorithm can produce a better segmentation, while processing with large-scale UAV images in a short time, indicating an excellent applicability under various conditions in the images of straw coverage. The finding can provide a promising potential way to improve the segmentation accuracy for the detection of straw coverage in modern agriculture.