CPDD: A Cascaded-Parallel Defect Detector with Application to Intelligent Inspection in Substation

Abstract

The intelligent inspection is a detection problem that aims to recognize abnormalities in substations. Defects acquired by various devices with small size, truncation, and similar appearance are easily confused, which biases the evaluation metrics. How to correctly explore the relationships between equipment and defects, and fully utilize results from different models is critical for this task. In this work, we propose a novel solution to these problems based on the cascaded-parallel defect detection (CPDD) algorithm. Specifically, it consists of two key components: (1) The cascaded model aims to mine the detailed relationships and filter out the illogical bounding boxes. This way can reduce the miss detection rate. (2) The parallel model is to fuse results from the mentioned cascaded model. It can utilize the information from these two-stage models and promote the detectable rate. Extensive empirical results on the dataset, acquired by our designed inspection system in different voltage-level substations, demonstrate the superiority of our proposed method. It can achieve state-of-the-art performance.