Target Detection Technique For Large-Field Optical Remote Sensing Image

With the rapid development of space remote sensing technology of earth observation,optical remote sensing imaging satellites are emerging,which makes images have the advantages of high resolution,wide coverage,rich details and so on.Image processing technology of remote sensing is also widely used in military defense and civil economic construction.Ships are the important targets of real-time monitoring and wartime attacks at sea,and their accurate and fast detection can play an important role in the analysis of enemy situation,precision guidance and military mapping.At the same time,this plays an irreplaceable role in rescue and the safety management of fishing vessels and so on.However,automatic detection is prone to false alarm and miss detection due to complex interference such as shooting weather,sea surface clutter,cloud,fog occlusion and uneven illumination.From the huge data,how to timely and reliably detect and extract the target becomes a crucial and urgent problem to be solved by ship detection in remote sensing image.In order to improve the accuracy and reliability of automatic detection and processing of ships in remote sensing images,to alleviate the pressure of transmission and storage data in-orbit,and to ensure the timeliness of remote sensing information,this paper has carried out in-depth research on the automatic detection of ship targets in visible large-file remote sensing image.This paper analyzes the imaging characteristics of remote sensing images,and designs two different detection algorithms for ships on the sea and inshore ships.And it also elaborates on the key techniques involved in the algorithm,including dehazing algorithm and sea-land separation method,target region of interest extraction and location,target feature extraction and false alarm elimination and bow positioning of inshore ships,etc.This paper improves the slow operation speed and low detection rate of ship detection algorithm under complex background in wide remote sensing image.In the image preprocessing stage,the optical remote sensing image is easily affected by the shooting weather and angle during the imaging process,and the fog appearing in the image interferes with target detection,resulting in the problem of missing information.Based on the atmospheric scattering model,this paper analyzes the problem of accurate estimation of the transmittance of the original guided filter,and proposes a dehazing algorithm based on the fusion gradient factor to more effectively maintain the edge and avoid excessive smoothing or insufficient smoothing.Experiment shows a good defogging effect is obtained.For the phenomenon that the dehazing video has inter-frame jitter and flicker caused by the motion of the camera or the target in the video sequence,on the basis of single-frame dehazing,this paper introduces the cross-correlation factor.A video dehazing algorithm is proposed to keep the continuity of frames,which reduces the inter-frame jitter after video dehazing.In the sea-land separation stage,this paper focuses on the problem of large error of sea-land separation by using the traditional gray-scale statistical method.The normalized water body index method is used to sea-land separation according to the principle of different reflectivity of different objects.And this method reduces interference of land false alarms.In the region of interest extraction,in order to avoid global search on the whole image,this paper proposes an extended wavelet transform,which enhances the contrast between target and background in complex background.The local maximum value search is performed and potential target points are located quickly.Then,based on the analysis of the principle of visual saliency model,an improved frequency domain hyper-complex saliency algorithm is proposed according to the characteristics of visible remote sensing images,and the saliency detection is performed within the local range of the suspicious points.Experiments show that this method is more effective in suppressing sea surface cloud,clutter,shadow and other interferences.In the phase of feature description and false alarm rejection,some non-target distractions will still be retained after salience extraction.In this paper,a 10-dimensional feature description operator is designed,which includes shape,texture and size information.SVM-based offline training is performed on a large number of sample images to obtain a stable training model,which is used to distinguish between ships and false alarms.Experiments show that the 10-dimensional characterization operator has strong ability to represent targets,and can effectively eliminate coastlines,islands,clouds,and reduce the false alarm rate.After the above steps,the whole process of ship detection on the sea area is completed.Then,a reliable and efficient detection method for the landing ship is proposed.Aiming at the difficult problem of inshore ship detection influenced by the surroundings,this paper proposes a method for detecting the inshore ship combing with the characteristics of the hull and the bow.Firstly,for the military port of interest,the template image is developed.And then edges cross-correlation algorithm under log polar coordinates is used to perform the port position of the real image.The "V"-shaped feature matching of the bow is carried out by using the improved shape context method.This method fully considers the problem of rotation and scale between the real shot bow and the "V"-shaped template due to different shooting angles and time.This method improves the accuracy of the match and can accurately identify potential suspicious ship targets.Then,according to the structure of the bow,this paper determines the direction of the hull by using line detection.Ship detection is confirmed combined with simple geometric features.In terms of comprehensive performance,this method can better achieve the positioning and detection of inshore ships.In terms of hardware development,this paper builds FPGA+DSP platform and optimize the ship detection algorithm.While ensuring the versatility and scalability of the system,the processing efficiency of the hardware is improved,and the transmission pressure of the remote sensing data on the star is reduced,which lays a foundation for the on-orbit target detection.