Underwater Mine Detection Using Histogram of oriented gradients and Canny Edge Detector

Abstract

For the naval defense forces, underwater mines pose a serious threat to safety and security to their lives and property. Studies conducted in this field which used side scan sonar imagery have not yielded sufficient accuracy for the detection of underwater mines and hence can lead to false alarms. In this paper feature extraction methods-Histogram of oriented gradients and edge-based feature extraction are used. These methods were chosen as they have shown very high accuracy in other studies which used different datasets. The data undergoes preprocessing-resizing and converting to grayscale images-after which the feature extraction method is applied. To classify whether the image contains a mine or not, template matching and classification methods feature vectors are used. It was found that this method yields high accuracy for the detection of mines. This same study can be extended for other object detection methods. The method followed here can help the naval defense in more accurate detection hence minimizing the damage which can be incurred in case of contact with a mine.