Research On Water Surface Target Detection Technology Based On Dynamic Optical Vision

To strengthen building a marine power and vigorously develop marine equipment must not only go to the oceans,but also respect the oceans,protect the oceans and reuse the oceans.This is one of the important contents of the 19 th CPC National Congress.As one of the underwater operation platforms,Unmanned Underwater Vehicles(UUV)have gradually increased its speed,range and endurance.It can carry weapons and detect targets remotely.It plays an indispensable role in the military field.Therefore,the research on water surface target detection technology based on dynamic optical vision has important military strategic significance and engineering application value.When the camera is installed on the UUV and leaned out of the sea,or installed on the swaying maritime buoy,visible light images of water surface are obtained.This paper mainly realizes the correlation processing of visible light images of water surface,and the detection and classification of the water surface ship targets.The main contents are as follows:(1)The characteristics of visible light images of water surface are analyzed in detail.Its background is complex,vulnerable to sea fog and other complicated weather,and the target appears mostly in the water boundary area.Then,from the two aspects of smooth denoising and degraded image enhancement,algorithm research and experiment are carried out to realize the preprocessing of visible light images of the water surface.(2)The background of the water surface images is complex.This paper studies a variety of water boundary extraction methods,and proposes an improved water boundary detection method based on the linear segmentation detection.This method not only adapts to the background of sea and sky,but also adapts to the background of waterfront.It is a real-time and adaptive algorithm.(3)The physical model of fog imaging and the classical dark channel priori fog removal algorithm are studied.However,the images of water surface contain bright areas,such as large areas of the sky and white waves of the water surface,and the dark channel priors are not always established.In this paper,the foggy images are divided into sky and non-sky regions by means of water boundary.An adaptive surface blurring method is proposed to calculate the transmittance,and then to obtain the atmospheric scattering function.At the same time,the improved method of quadtree subdivision is used to estimate atmospheric light.Through experiments with a variety of defogging algorithm to do a comparison,quantitative and qualitative analysis.(4)The water surface target detection is realized,and then the classification of ship targets is further realized.With the help of two binarezed normed gradient and Surpport Vector Machines training detection template,the target region of interest and its position coordinates are obtained.The SURF feature points on the image of the target region of interest are extracted and characterized.As input information,a visual bag model classifier is trained to classify ship targets.