基于M_SFANet算法的羊群密集计数系统设计

Design of sheep dense counting system based on M_SFANet algorithm

为了更好地开展过度放牧监测和草原生态评估工作,文中针对当前羊群计数系统存在的弊端,设计一种羊群密集型计数应用系统。该系统包括羊群密集计数算法和应用系统设计两部分。羊群密集计数算法部分借鉴人群密集计数模型M_SFANet,在自建羊群密集计数数据集USC(UAV Sheep Counting)上进行计数,通过合理处理整个输入图像中目标对象的尺度变化,得到质量较好的预测密度图;再对预测密度图进行积分得到羊只数量。结果表明,M_SFANet模型得到的密度图平均绝对误差(MAE)、均方误差(MSE)和计数平均相对误差(MRE)分别为3.61%,5.44%和1.87%,计数精度较高。此外,基于M_SFANet计数模型,在Ubuntu系统上建立一个Web服务器,首先采用Bootstrap、HTML5将前端数据进行可视化;然后使用Flask框架进行后台开发服务,同时融合MySQL数据库技术完成密集计数应用系统的设计,从而实现数据传输到应用的完整功能,使得用户可在客户终端浏览计数系统网页,并通过此网页进行计数及查询历史记录。相对于其他羊群计数方法,文中所设计系统具有较高的计数精度,且方便快捷、节约成本,发展前景更加广阔。

In allusion to the defects of the current sheep counting system,a sheep dense counting application system is designed to better conduct overgrazing monitoring and grassland ecology assessment. The system is composed of the sheep dense counting algorithm and the application system design. In the sheep dense counting algorithm,the crowd dense counting model M_SFANet is used for reference to count on the self-built sheep dense counting dataset of USC(UAV sheep counting),the predictive density map with good quality is obtained by handling the scale variation of target objects in the whole input images reasonablly,and then the predictive density map is integrated to obtain the number of sheep. The results show that the mean absolute error(MAE),the mean square error(MSE)and the counting mean relative error(MRE)of the density map obtained by the M_SFANet model are 3.61%,5.44% and 1.87% respectively. The counting system has relatively high counting precision.Based on the M_SFANet counting model,a Web server is built on the Ubuntu system. The Bootstrap and HTML5 are used to visualize frontend data, the Flask framework is used to develop the backend services, and meanwhile MySQL database technology is fused to complete the design of the dense counting application system,so as to realize the complete function of transmitting the data to the application,which can facilitate users to browse the counting system web pages on client terminals and execute counting and query history records. In comparison with other sheep counting methods,the designed system has higher counting accuracy,is convenient,efficient and cost-saving,and also has a broader development prospect.