Small insulator target detection based on multi‐feature fusion

• Published in: IET image processing Vol. 17; no. 5; pp. 1520 - 1533
• Main Authors: Tang, Minan, Liang, Kai, Qiu, Jiandong
• Format: Journal Article
• Language: English
• Published: Wiley 01.04.2023
• Subjects: feature extraction; flaw detection; insulators; support vector machines
• ISSN: 1751-9659; 1751-9667
• DOI: 10.1049/ipr2.12735

The proportion of insulators in aerial power patrol images is small and the background of overhead lines is complex, often leading to incomplete and inaccurate detection of insulators. Therefore, an algorithm for detecting insulator targets based on multi‐feature fusion is developed in this study. Firstly, a dynamic threshold oriented fast and rotated brief algorithm is proposed, which uses the bag‐of‐words dictionary model to determine local shape features of the image, applies gradient weighting to the global texture feature vector extracted by the histogram of oriented gradients algorithm and performs radial gradient transformations to get the improved HOG of features. Secondly, the feature vectors are fused serially, the learning machine is trained and the parameters of the support vector machine are optimized using the quantum particle swarm optimization algorithm. Finally, the target area is pre‐divided by the selective search algorithm, and the area is classified by the learning machine. The experimental results show that the proposed feature extraction method can describe the image details more accurately than the existing methods, and the average accuracy of the feature extraction classifier can reach 93.7%, which helps to overcome the incomplete detection problem of insulator detection at the aerial work site. A dynamic threshold ORB algorithm is presented, which combines BoW models to form local feature operators. A HOG global feature extraction algorithm based on RGT variation is proposed. SVM parameters are optimized by QPSO algorithm.