Abstract: Aiming at the problem of poor detection effect caused by the large difference in the scale of detection targets in aerial insulator images and insulator defects with small scale and complex background, we put forward a multi-defect detection algorithm for aerial insulators based on improved YOLOv7. The algorithm uses a concat bidirectional feature pyramid network (CAT-BiFPN) instead of path aggregation network (PANet) in YOLOv7 to reduce the redundancy in the structure aiming at fusing different features, to improve the fusion of multi-scale target features, and to form the 4th detection layer targeting the detection of small targets. The algorithm further distinguishes between different insulator defects by adding a mixed model of self-attention and convolution (ACmix) that pays more attention to the details in the features. The algorithm detects normal insulators, self-explosions, fouling and breakage on high voltage transmission lines in aerial images and simultaneously detects bird's nests on towers as foreign objects. The experimental results show that the mean average precision of the algorithm in this paper is 93.9%, which is 9.6% higher than that of the standard YOLOv7, and the defect detection algorithm proposed in this paper can better identify defects in insulators of different scales accurately.