Improving the Accuracy of Frosting Image Recognition for Cold Storage Evaporators Based on Gray-Level Co-occurrence Matrix and Multi-layer Extreme Learning Machine
Haiyang Zheng, Shan Jiang, Feng Wang, Xin Wang · 2024
Currently timed defrost and on-demand defrost are the two main defrosting methods in the refrigeration industry. As one of the on-demand defrosting methods, the use of image recognition technology for on-demand defrosting of cold storage heat exchangers is considered to be an efficient and easy-to-operate control method. However, for different operating environments in cold storage refrigeration systems, the recognition accuracy of the heat exchanger defrosting state is greatly reduced by the existing digital image processing techniques. For this reason, a method for recognizing the evaporator frost state using a gray scale covariance matrix combined with a multilayer limit learning machine (GLCM-ML-EML) is proposed. Using the laboratory environment, a large number of pictures of cold storage evaporator frosting under different operating environments were taken. And the frost pictures were categorized into three categories (no frost, light frost, and heavy frost). The experimental results show that in ternary classification, the recognition accuracy is up to 98.84% based on the gray-scale covariance matrix with multilayer limit learning machine. The recognition accuracy is improved by 6.77% compared to the traditional CNN model. It shows that this technique has high recognition accuracy and has good application prospects in the field of cold storage defrost control.