Research On Weak And Small Ship Target Detection In Images

In the process of developing the oceans,disputes between China and neighboring countries over maritime resources rights and interests occur from time to time.It brooks no delay to protect marine rights and interests for the main channel of our economy and important source of production.Ship is an important tool for the development of the ocean.How to detect ship targets accurately is the key to exploit the sea.At present,the satellite-borne visible light remote sensing platform is one of the important mean to detect ship target,but the proportion of the target area is small,the target features are weak and the shape and texture are not obvious.How to detect the weak and small ship targets in the remote sensing image has become current research hotspot.This thesis will focus on the detection of weak and small ships in satellite-borne visible light remote sensing images.The main tasks are as follows:(1)A detection method of ship weak small target based on DFSNDLN(Decision Fusion of Shallow Networks and Deep Learning Networks)is designed.The main idea of this method is as follows: Firstly,the method based on traditional shallow network has a low false alarm rate and a high false alarm rate,while the method based on deep learning has a high false alarm rate and low false alarm rate.Secondly,the decision tree is used to combine the results of two detection methods which contains the tradition shallow network results and the deep learning network,so as to realize the detection of ship weak small targets with high accuracy,low false negative rate and low false positive rate.(2)Experimental verification on the self-built data set of small and weak ship targets was carried out The results showed that the accuracy rate of the weak and small ship target detection method based on HOG+SVM is 74.7%,the missed detection rate is 7.8%.Meanwhile,the accuracy rate based on improved YOLOV5 method detection network for weak and small ships is 89.0%,the missed detection rate is 8.9%.Otherwise,the detection method for small and weak ships based on DFSNDLN has an accuracy of 89.2%.The missed detection rate is 7.0%.To sum up,the method proposed in this paper can achieve the detection of weak and small ships with high accuracy,low false alarm rate and low false alarm rate.(3)Base on the detection method which is designed,a detection prototype software for small and weak ship detection was developed by comprehensively using tools such as Python,Pytorch and Py Qt.Meanwhile test and verify the software.