Research On Key Techniques Of Maritime Target Detection In Visible Bands Of Optical Remote Sensing Images

In recent years,with the development of the earth observation technology,remote sensing imaging has entered an unprecedented new stage.A large number of the high-resolution optical remote sensing satellites have emerged and can provide submeter-resolution panchromatic images.In addition,aerial remote sensing,such as UAV(Unmanned Airborne Vehicles),can also provide high-definition images in the near-earth.Maritime target detection and recognition are provided with very rich data source by astronautics and airborne remote sensing technology,and become an active research field.It plays a crucial role in a spectrum of related military and civil applications.As important targets for marine monitoring and wartime strike,detection and identification of ships can master the enemy's operational strength.The distribution of ships can be monitored in key sea areas,and the intelligence of the naval operations can be analyzed.Besides,precise guidance can be carried out and naval defense and security can be ensured.In civil field,ship detection also holds the key for a wide array of applications,such as traffic surveillance,maritime rescue,protection against illegal fisheries,anti-smuggling efforts,oil discharge control,and sea pollution monitoring.Therefore,ship detection is important to protect the coastlines and explore the rich marine resource.However,there are still plenty of difficulties in this field.For instance,a ship's appearance may vary greatly due to uneven illumination,the variability of the ship sizes and the viewing geometry.In addition,the sea surface is complex due to the interference from clouds and haze,sea clutters,ship wakes,small islands,and coastlines,which may be falsely detected as ships.The false alarms may be produced and the processing time may increase.Therefore,determining how to accurately,quickly and stably detect ship targets in the marine background,and win response and processing time as more as possible is an urgent problem.In order to improve the efficiency,accuracy and reliability of remote sensing images data processing,a plenty of studies have been made.In this paper,we focus on the automatic detection of ship targets in the visible bads of the remote sensing image.We have studied the key technologies about ship target region extraction,image segmentation,and target discrimination.On the base of these,a series of automatic detection methods for maritime targets are designed with high detection precision and low false alarm rate.The efficiency and adaptability of the detection system for ship targets are also improved.The innovative research work and results of this paper are summarized as follows.1.Recently,it has become well known that the visual saliency model can quickly access to information associated with current scene and task,even for a highly cluttered scene.This advantage has made it a hot spot in ship detection.In this paper,the significance of the visual saliency model is discussed.The construction of the visual saliency model and its application in the ship target detection in the visible bands of the remote sensing image are explored in depth.In addition to achieving higher accuracy,a visual saliency model is designed in this paper.We have improved the existing models and constructed a practical combined saliency model for further,which integrates multi-frequency information using self-adaptive weights based on Entropy information.It is effective in extracting both of the large ship and small ship,and the target region is not complete and not uniform.Most of the thin clouds,mist and sea clutter are removed and the interference from complex backgrounds is suppressed.In addition,our model is not sensitive to the parameter settings and the whole method can be automatically executed.2.After the target detection based on the visual saliency model,some pseudo-targets,such as thin clouds,sea fog,shadow and sea clutter,can be inhibited.But the heavy clouds,islands or coastlines and other distracters may be still included.To further reduce these false alarms,other techniques are needed to effectively remove the interference according to the characteristics of the ship and non-ship targets.The shape feature of a ship is more regular since it appears as a long symmetrical strip,whereas the shapes of the pseudo-targets detected are irregular.Inspired by this fact,a novel descriptor is designed to identify real ships based on gradient features.It eliminates the interference from the false targets,such as thick clouds,islands,coastlines.The real ship targets are retained.The method achieves robustness against scenes with clouds,island and clutter and is effective in the presence of ship size variantion and ship wakes.The ship detection in the ocean surface of the optical remote sensing image is realized from coarse to fine.3.Recently,owing to the characteristics of the multi-scale and multi-direction wavelet analysis,Wavelet transform is gradually valued by researchers in the saliency modeling.Ship target is viewed as uncommon regions in the sea background caused by the differences in colors,textures,shapes,or other factors.Inspired by this fact,a global saliency model based on wavelet transform is designed to improve the adaptive performance of the ship detection on the different resolution sea images.Wavelet coefficients can express them effectively based on the spatial-frequency analysis of the wavelet transform,which can examine the signal at different bands and bandwidths.Considering the fact,the factors including different colors,different scales,and different orientations are used for simulating our saliency model.A series of feature maps with the high-frequency data can be created by inverse wavelet transform operation.They are constructed based on high-frequency coefficients of the multi-scale and multi-direction wavelet decomposition.They can characterize different feature information from the edge to the texture of the input image.The global saliency map is calculated by the Gaussian probability density function.Then,the distance decay formula is implemented to weaken the non-salient information in the saliency map.The large range low-frequency information from the sea background can be suppressed,while most of ship regions with clearer contours can be extracted accurately.The ship regions extracted is uniform and complete.In addition,the images with different sizes can be effectively processed by the proposed saliency model.4.In some cases,the sizes of some targets in remote sensing image is too small to fine segmentated.Aim at this problem,a new and effective multi-level discrimination method is designed for further false alarms reduction.According to the characteristics of the ship and no-ship chips,the improved entropy estimation is presented,which overcomes the deficiency of the traditional entropy relying on spatial geometric information.The structure information and the pixel distribution of the ship and the non ship chip images are taken full consideration in this step.It can remove the false candidates and retain the real ship targets.For the false chips whose entropy values are confused with those of the ship targets,the pixel distribution identification is proposed for further decreasing the false alarm rate.In this multi-level discrimination method,some image preprocessing work,such as fine segmentation and feature extraction,are not needed.Owing to these novel techniques and improvements,the presented approach shows higher discriminative power.After the target identification,the locations and the number of the ships in various sizes and colors can be detected accurately and fast with high robustness.Overall,the related theories and challenges faced in the field of ship automatic detection in the visible bands of optical remote sensing images are introduced and analyzed.Some key technologies,including the improvement of the visual saliency model and the construction of the deep learning network,are studied and analyzed in depth.Some results are acquired in the study.The relevant achievements of this paper provide the theoretical foundation and algorithm support for the ship automatic detection in the visible bands of the remote sensing image.These solutions can also be applied to other remote sensing target detection system and they are of reference significance.