Research On Infrared Ship Object Detection Method

As an important carrier of maritime military forces,naval ships are critical objects for military surveillance and strikes.Effective detection and identification of ship objects in complex maritime environments have become a hot research topic in the military field both domestically and abroad.Currently,infrared ship object detection and recognition algorithms still face issues such as low detection rates and poor adaptability under interference from weather and island/terrain backgrounds.This paper address the practical needs of the field and study the infrared ship object detection and recognition algorithm based on saliency detection and convolutional neural network technologies,from both traditional algorithm and deep learning algorithm perspectives,to achieve effective detection of infrared ship objects under the interference of various complex sea and sky backgrounds.The main research contents of this paper are as follows:(1)A method is proposed for infrared ship object detection and recognition based on saliency detection algorithms and invariant moments.A denoising and enhancement algorithm is designed for infrared ship object images,use saliency detection algorithms and efficient subwindow search to extract salient regions of the image,and then comprehensively recognize the ship objects by extracting their geometric features,statistical features,and invariant moments.The experimental results show that the proposed algorithm achieves a comprehensive detection and recognition rate of over 90%,and can complete automatic detection and recognition of ship objects under complex sea and sky backgrounds.(2)An infrared ship object detection algorithm is proposed based on an improved CenterNet algorithm to improve the algorithm’s detection rate and anti-interference performance in complex island/terrain and sea/sky scenarios.The Mosaic data augmentation algorithm is introduced for data enhancement and the Dilated Encoder module and 3×3 null convolution are added to optimize the CenterNet backbone network,enhance the fitting performance of the network and improve recognition accuracy.This algorithm achieves excellent detection accuracy for ship images with island/terrain backgrounds.The optimized CenterNet+ algorithm can achieve an accuracy of 98%,a 6% improvement over the original algorithm,while maintaining the original algorithm’s speed,effectively achieving infrared ship object detection.(3)A multi-label infrared image scene perception grading algorithm is proposed based on weakly supervised learning,which introduces multi-label classification algorithms into the field of scene perception for infrared multi-scenario and multi-object images.This paper construct an infrared multi-scenario and multi-object dataset,use the ConvNeXt backbone network to complete the feature extraction task,and introduce spatial attention mechanism algorithms and asymmetric loss functions to improve the adaptability and accuracy of multi-label classification algorithms.(4)This paper construct an infrared ship object detection system,which uses the multi-label classification grading algorithm for infrared ship images to provide important scene information for object detection.The object detection model can adaptively use saliency detection algorithms or the CenterNet+ deep learning algorithm to detect and recognize image objects in different grading scenarios,meeting the practical needs of the field and having high application prospects and significance.