Abstract: A lightweight convolution neural network (LightweightES) for edge computing equipment is proposed to identify electrical equipment and their abnormal heating faults in thermal images. In order to reduce the number of model parameters and improve detection accuracy, the classical SSD was modified as follows. MobileNetV3 lightweight network was used as the backbone network of feature extraction to extract image features efficiently. The efficient channel attention module (ECA) was introduced to improve the detection accuracy of the network. The SoftPool method was used to reduce the loss of the pooling information and improve the classification accuracy. A data set of 10516 labeled infrared images of electrical equipment was established including 6 types of outdoor substation equipment, such as current transformers, arresters, insulators, disconnectors, circuit breakers and drivepipes. The experimental results show that the mAP of LightweightES algorithm reaches 93.8%, which is 7.5 percentage points higher than SSD. The number of parameters is only 1/5 of SSD, while the detection frame rate is up to 55 FPS, which can accurately identify the electrical equipment and local temperature abnormal faults in real time. It is suitable for intelligent field monitoring terminal with limited computing power.