Research On Computing Technology Of High Performance Image Fusion Guidance

The performance of missile guidance systems directly affects the hit accuracy of missiles on mobile targets.In order to improve reliability and robustness of the guidance system,the multimode complex guidance method has been widely used.Radar-based guidance system needs to transmit the detection signal to track the target.It is easy to be found and interfered by the anti-missile guidance methods,reducing the hit accuracy of the missile.Infrared images or visible image-based guidance systems can use the spectral signal of targets,without transmitting track signals to the targets,therefore they are good for hiding and have been a popular approach.However,infrared image or visible based guidance systems have their own inherent drawbacks.Infrared images can observe hot targets under poor illumination and smoke obstruction with all-day features,but generally the results are low contrast and lack image details.Visible image methods have high contrast and resolution with better edge and texture details,but are vulnerable to factors such as light intensity and smoke interference.Therefore,it is critical to guide missiles based on the fusion of infrared image and visible image to utilize the advantages of both and improve the performance of the missile guidance.To integrate infrared image and visible image,we must first solve the system architecture problems,that is,how to design an architecture to maximize the advantages of using both images;Secondly,since the main purpose of the image fusion is for better target recognition,normal evaluation and integration methods for image fusion cannot effectively deal with the fusion problem in missile guidance;Thirdly,uncertainty exists in the target image tracking process due to obstruction,interference and other unknown conditions.Therefore,the system must have the ability of anti-interference;In addition,image processing,especially multiple images fusion for target recognition,requires strong computing capabilities.Unfortunately,the existing methods based on serial computing are difficult to meet the demand.This thesis studies current work and challenge of research on the image fusion for missile guidance.It also presents in-depth analysis for a number of key technologies on the image fusion for missile guidance.The major contributions include:1.Because there are a large amount of image information loss on decision level fusion and little information loss on data level fusion,we have designed and implemented the multi-source image double multimode tracking architecture to solve their problems.The architecture first runs the pixel-level fusion for the visible and infrared images,and then uses a variety of ways to identify the targets in the three images.Finally it performs the decision-making level fusion for multiple recognition results,in order to overcome disadvantages in the tracking modes with single image and single identification,so that it can improve the robustness of target recognition and tracking.For systems that require high performance computing features,we deploy embedded CPU / GPU heterogeneous parallel computing architecture to design related hardware system architecture and multi-layer software architecture that can be easy maintained,based on a high-performance low-function So C chip(integrate CPU and GPU).2.The traditional pixel-level image fusion methods and evaluation methods are not suitable for the missile guidance applications based on the identification of target segmentations.Therefore,we have proposed an image fusion evaluation method based on maximum histogram curve variance ratio,and a fusion algorithm based on the salient features of the image brightness.The approach using maximum histogram curve variance ratio can evaluate if the image foreground and background can segment targets.Experiments show that this method can better evaluate if the fused image is suitable for the identification of target segmentation for the missile guidance applications.The fusion algorithm using the salient features of the image brightness makes the brightness of target features as main considerations,and takes discrete wavelet transform to integrate two images by weighted fusion.Experiments show that the fusion algorithm is suitable for applications in object segmentation and recognition.3.Considering that there are uncertainty such as target obstruction and interference in the target image tracking of missile guidance,we have incorporated the ideas from fuzzy systems,and designed and implemented a fusion algorithm for target identification results,based on the T-S fuzzy fusion systems.The algorithm processes target identification for each image using the active contour tracking method and the target area tracking method,and then filter every identification result by the H? filtering.Finally,it uses T-S fuzzy system to integrate a variety of identification results.Experimental results show that using H? filtering and T-S fuzzy system for data fusion can effectively solve the target obstruction and interference problems in the single source image and single mode identifications.4.In order to improve the computing performance in our system,we have designed and implemented parallel GPU acceleration for image fusion guidance systems.We have analyzed the system parallelism,designed parallel strategy,and complete parallel computing to accelerate the processing of main images by optimized implementation.Experimental results show that the algorithm using GPU parallel acceleration can improve processing speed significantly.