Comparative study on adaptable intelligent frost recognition method for air-source heat pump and cold chain based on image texture features under complex lighting conditions

Abstract

The energy efficiency enhancement of refrigeration/heat pump systems is a crucial aspect of carbon emissions reduction. Accurately recognizing the frosting state of their evaporators in low-temperature environments to achieve precise defrosting is key to reducing system energy consumption. Intelligent recognition methods based on evaporator images hold promise for high recognition rates. However, in practical conditions, light intensity can severely reduce the identification accuracy of existing methods, necessitating improvements. Therefore, a highly adaptable new method based on texture features of evaporator surface images is presented in this study, where texture features is extracted by minimum-redundancy-maximum-relevance-enhanced gray level co-occurrence matrix, and classified by sparrow-algorithm-optimized extreme learning machine (GLCM-SELM), to overcome the impact of various light intensity. This method is validated using a dataset of 4125 evaporator images of three frosting states, which is experimentally collected under light intensity ranging from 5 to 2370 lx. Performance study and comparative analysis against existing methods are carried out. Results indicate that the new method achieves identification accuracy of approximately 95 % across different conditions, significantly outperforming existing methods by 6 % to 35 %. Its remarkably smaller standard deviation (0.05) demonstrates high stability. It also shows fast computing speed and low cost. Generally, it has good application potential.