基于语义分割的冷库排管霜层厚度测量研究

Research on Measurement of Frost Layer Thickness in Cold Storage Pipe Arrays Based on Semantic Segmentation

摘要: 针对冷库霜层厚度监测过程中常处于低光照、多障碍的环境，严重影响图像质量与霜层厚度测量精度的问题，提出了一种基于语义分割的冷库排管霜层厚度测量方法。首先将语义分割算法引入图像处理中对采集所得的低光照霜层图像进行像素分割，以消除低光照环境对图像质量的影响；其次对排管霜层的结霜区域进行Canny边缘检测，并结合霍夫直线检测算法，实现对霜层区域的提取；最后利用冷库排管的实际尺寸与结霜区域像素宽度比例关系确定霜层厚度。实验结果表明，文中所提方法能够有效计算低光照、多障碍环境下的冷库排管霜层厚度，计算所得的霜层厚度相对误差仅为1.88 mm，对冷库排管霜层监测工作具有一定的参考价值。

Abstract: A semantic segmentation based frost thickness measurement method for cold storage pipes is pro-posed to address the problem of low light and multiple obstacles in the process of monitoring frost thickness, which seriously affects image quality and frost thickness measurement accuracy. Firstly, the semantic segmentation algorithm is introduced into image processing to perform pixel seg-mentation on the collected low light frost layer images, in order to eliminate the impact of low light environment on image quality; Secondly, Canny edge detection is performed on the frosted area of the exhaust pipe frost layer, and combined with the Hough line detection algorithm, the frost layer area is extracted; Finally, the frost layer thickness is determined based on the proportional rela-tionship between the actual size of the cold storage duct and the pixel width of the frosting area. The experimental results show that the proposed method can effectively calculate the frost layer thick-ness of cold storage pipes in low light and multi-obstacle environments. The relative error of the calculated frost layer thickness is only 1.88 mm, which has certain reference value for cold storage pipe frost layer monitoring work.