一种高精度的光伏阵列红外图像分割算法-《计算机技术与发展》

文章信息/Info

Title:: A High-precision Photovoltaic Array Infrared Image Segmentation Algorithm

Author(s):: CHEN Wen-qin; HAO Hui-jie; XIAO Jian; GUO Yu-feng; School of Electronic and Optical Engineering,School of Microelectronic,Nanjing University of Posts and Telecommunications,Nanjing 210023,China

Keywords:: infrared image; photovoltaic array; image segmentation; local standard deviation; high precision

摘要:: 随着化石能源短缺,且其燃烧带来的环境污染问题,清洁的太阳能发电得到更广泛的使用。然而光伏阵列长期处于恶劣环境中,导致光伏组件容易出现损坏。传统光伏故障检测方法效率较低、准确率不高,难以满足实际需求。因此在分析光伏阵列图像特征的基础上,提出基于光伏阵列区域局部灰度特征的高精度分割算法。通过使用? 5×5 的滑动窗口计算 Gaussian 处理后的灰度图像的局部标准差,以此衡量局部灰度一致性,提取出灰度一致性高的区域。应用形态学膨胀处理所得区域,并使用基于尺寸的筛选方法去除孤立小区域,得到最终分割结果。实验结果表明,该算法对于光伏阵列的识别准确率达到了 96% 以上,同时降低了背景区域的误检率。算法可精确、有效地分割出光伏阵列区域,分割效果优于边缘检测和 Otsu 算法。

Abstract:: With the shortage of fossil energy and the environmental pollution caused by its combustion,clean solar power has been widely used. However, the photovoltaic array has been in a harsh environment for a long time,causing the photovoltaic modules to be easily damaged. Traditional photovoltaic fault detection methods have low efficiency and low accuracy,which is difficult to meet actual needs.Therefore,based on the ana-lysis of the photovoltaic array image characteristics,we propose a high-precision segmentation algorithm based on local gray features of the photovoltaic array region. By using a 5×5 sliding window to calculate the local standard deviation of the gray image after Gaussian to measure local gray consistency,the regions with high gray consistency are extracted. Apply the morphological expansion process to the region obtained, and propose a size-based filtering method to remove the isolated small regions to obtain the final segmentation result. The proposed algorithm achieves a recognition accuracy of more than 96% for the photovoltaic array,while reducing the false detection rate of the background region. The algorithm can accurately and effectively segment the photovoltaic array region,and the segmentation effect is better than traditional algorithm,such as the edge detection and Otsu algorithm.