Robustness analysis and performance comparison of a novel intelligent frost identification methods for air-source heat pump based on gray-level Co-occurrence matrix

Air-source heat pump plays a crucial role in the global energy transition as efficient heating solutions. However, frost accumulation on outdoor evaporators significantly degrades their performance in cold climates. Currently used defrosting methods suffer from low accuracy (≤85%) and frequent false defrosting (up to 68%), particularly under varying lighting conditions and viewing angles, leading to substantial energy waste. To address this, a novel intelligent frost identification method based on the Gray-level Co-occurrence Matrix combined with the Sparrow Search Algorithm and Extreme Learning Machine (abbreviated as GSE) is proposed. This study systematically investigates the performance of the GSE method through experiments capturing evaporator surface images under three frost states (frost-free/light/heavy) with different lighting intensities (4-1000 lux) and shooting angles (left/front/right). The proposed GSE method achieves excellent frost recognition performance under variable lighting and angles, with an average accuracy of 97.79%. Compared to traditional BP and ELM models, it offers higher accuracy, better robustness, and faster recognition (average 1.413 ms per image). Therefore, GSE can facilitate intelligent defrost control in air-source heat pumps, thereby improving energy efficiency and contributing to carbon neutrality goals.