适应场景光照变化的桔小实蝇诱捕监测系统优化设计与试验

为了减少光照造成的桔小实蝇误判,该文在前期设计的基于视频监控的桔小实蝇诱捕装备的基础上,优化改进了硬件装备及监测执行算法。主要通过对装备遮光系统和太阳能装置的改良及Wi Fi/4G等技术的应用形成了一套较完善的集采集、处理、传输和供电为一体的桔小实蝇诱捕监测新装备。同时,改进提出了一种可适应场景光照变化的桔小实蝇检测算法(Bactrocera dorsalis detection algorithm under lighting variations,BDDA-LV),并对软件系统的控制流程与感兴趣数据的判别、获取及传输进行优化,最终形成了一套有较好人机交互的桔小实蝇监测软件系统。通过对新旧桔小实蝇检测算法及系统分别开展了室内和野外试验,结果表明,在中度光照影响下,BDDA-LV算法错误率为7.21%,比原算法降低19.07%,且其运算耗时为原算法的41.7%。在严重的光照影响下,BDDA-LV算法错误率为12.4%,比原算法降低23.07%,且其运算耗时为原算法的42.2%。同时在广州市天河区杨桃公园自然环境下进行了的较长期系统监测试验,该系统可实时稳定地对桔小实蝇进行监测计数。通过对2015年6月18日至24日自然环境下的试验,用计算机系统记录和人工计数比较分析,得到计算机系统计数1634头,人工计数1613头,准确度达98.7%,比优化之前在实验室进行监测试验的系统跟踪精度提高了3.8%。该文所提出的BDDA-LV准确率和运算速率都远高于BBDA算法,整个系统拥有较高的精度和稳定性,具有较好的推广应用价值。

Optimization design and experiment of monitoring system for Bactrocera dorsalis trapping under lighting variations

Bactrocera Dorsalis is a type of invasive pests and causes serious damage to many important economic crops. In order to monitor Bactrocera Dorsalis accurately, and to reduce the misjudgment of light-induced problem, a remote monitoring system which combined a hardware equipment and a software system was designed. The hardware equipment was optimized based on our preliminary design for Bactrocera Dorsalis trapping. By improving the shading system, solar devices and 4G communications devices, the hardware equipment now can be self-powered and integrate the insect pest information collection, processing and transmission as a whole functionality. In addition, this paper presents a good human-computer interaction software system, which includes the Bactrocera Dorsalis monitoring programs, the server and the client-side. The monitoring program can monitor the trapping process, precisely calculate the number of Bactrocera Dorsalis, and automatically transmit the results to the remote server or store it in a local storage card. Users are convenient to obtain the monitoring information through the client-side. In order to improve the accuracy of the Bactrocera Dorsalis detection algorithm (BDDA) and achieve the requirements of real-time systems, an algorithm named Bactrocera Dorsalis detection algorithm under lighting variations (BDDA-LV) was proposed. The BDDA-LV began by selecting appropriate background model according to the light conditions. Then the background difference method was used to extract Bactrocera Dorsalis target preliminarily. Median filtering and morphological filtering for the image was used to reduce white noise, and eliminate holes in the target area to improve the image quality. The image was then divided into blocks based on the adjacent pixels of the image and used these blocks for Geometric feature matching, so that the Bactrocera Dorsalis area can be extracted. Both BDDA-LV algorithm and BDDA algorithm were tested by two sets of data. These data came from the pre-acquired image dorsalis, and was divided into two datasets, one of which was under the moderate influence of light and the other one was under the severe influence of light. Each dataset had 150 images. In the influence of moderate lighting variation, the error rate of BDDA-LV algorithm was 7.21% and 23.07% lower than BBDA algorithm, and its computing time was 41.7% of the BBDA algorithm. In the influence of strong lighting variation, the error rate of BDDA-LV algorithm was 12.4% and 23.07% lower than the BBDA algorithm, and its computing time was 42.2% of the BBDA algorithm. By the long-term tests of the system in the Guangzhou Yangtao Park, the system ran stably and no power outage occurred. A comparative analysis of monitoring system records and artificial counting were carried out from June 18 to 24, 2015. The monitoring system counting was 1634 while artificial counting was 1613. The accuracy of the monitoring system was 98.7%, which improved 3.8% than the preliminary system. The experiments showed that the accuracy and efficiency of the BDDA-LV were higher than the original BBDA algorithm, and the whole system had higher precision and stability. This system can also easily provide accurate pest monitoring information to the regional monitoring personnel in real time, which improves work efficiency. At the same time, the system can provide information support for agricultural researchers to study on the insect activity. This system is thought to be valuable and applicable.