This paper aims to address the issue in existing object detection models that face a gradual attenuation of features during forward propagation from Backbone to Head

leading to insufficient feature representation

and to overcome the limited ability of traditional feature pyramid network (FPN) to effectively learn discriminative features

which hinders the model’s capacity to distinguish between insulator defects and the interference of similar background in complex environments. Moerover

a multi-level compensation 3D perceptual enhancement algorithm for insulator defect detection (MC3DPE) is put forward. Specifically

the multi-level detection head feature differentiated compensation (MDFDC) framework employs a gated activation mechanism to dynamically fuse frequency component information

enabling the correction of missing high- and low-frequency information at different levels of detection head

thus improving detection accuracy for insulator defects. The 3D perceptual reutilization progressive enhancement (3DPRPE) module constructs a 3D weight matrix through a space-channel feature interaction mechanism and adopts a weight reuse strategy to progressively learn multi-scale receptive field features

thereby improving feature discrimination capabilities. Experimental results show that

on our constructed insulator and defect detection dataset

the proposed algorithm achieves a mean average precision of 89.0%

surpassing the existing state-of-the-art 10 object detection algorithms by at least 2.7%. The algorithm provides valuable guidance for enhancing the recognition of insulator defects in transmission lines based on UAV imagery.