The Key Technologies Study On Machine Vision With Swarm Intelligence

Based on a number of advanced imaging devices and imaging technologies, machine vision is widely used in medicine, astronomy, defense industry and scientific research domain, etc. The machine vision system is divided into image acquisition, image processing and motion control, and it is referred to information science, computer science, mathematics, physics, biology and some other disciplines. Swarm intelligence (SI) optimization algorithm is one of evolutionary computational methods which simulate collective behavior of biology developed rapidly in recent years. Particle Swarm Optimization (PSO) algorithm is one of the typical SI algorithms. Quantum-behaved Particle Swarm Optimization (QPSO) algorithm is proposed based on in-depth research of PSO algorithm. Since this method involves only elementary operations, it has simpler optimization process, fewer parameters and faster convergence speed than original PSO algorithm. The improved QPSO algorithm is adapt to apply in machine vision system so as to enhance the flexibility.Image processing and analysis is the hard core of machine vision system. Swarm intelligence optimization algorithm is introduced to the key image processing techniques of machine vision system in this paper. Several image processing algorithms based on improved QPSO have been proposed, such as image enhancement, image restoration, image segmentation and target recognition. The above-mentioned algorithms have achieved positive results by using the parallel searching features of QPSO algorithm, and some of them have been successfully introduced into a high-speed online foreign fiber detection system with machine vision. The proposed foreign fiber detection system did improve foreign fiber detection rate and enhance system stability. The main contents are outlined as follows.(1) Image enhancement and image restoration is introduced to solve image degradation of machine vision system. Compared with traditional image enhancement techniques can only accommodate particular images, an adaptive image enhancement method based on improved QPSO algorithm is proposed. Simulation experiments through standard test functions show that the improved QPSO algorithm proposed in this paper has better convergence. The image enhancement experiment shows that the proposed method with swarm intelligence algorithm has better performance and adaptability. As some non-linear image restoration algorithms, for example the Lucy-Richardson algorithm, generally require significant computational resources, an improved image restoration method based on improved QPSO algorithm is proposed, and it can effectively reduce time complexity of image restoration.(2) Image segmentation is the premise stage of image analysis and recognition in machine vision system. Since minimum error threshold method for image segmentation can not use spatial information of input images, this article will promote the above-mentioned method from one-dimensional to two-dimensional space. Firstly QPSO algorithm with multi-group and multi-stage improvement is proposed, and secondly it is applied to optimization process of two-dimensional minimum error threshold segmentation method, lastly image segmentation is achieved by using a pair of thresholds from the previous step. The emulational experimental results through standard test functions show that the improved QPSO algorithm has better convergence. Finally, the proposed image segmentation method based on improved QPSO algorithm is compared with other image segmentation technique, and a large number of experiments show that the proposed algorithm has better image segmentation effect and stability.(3) Image classification and object recognition is one of the ultimate goals of machine vision system. In this paper, a new Support Vector Machine (SVM) classifier training method is proposed based on SI optimization algorithm, and it is applied to target recognition. Firstly Quadratic Programming (QP) problem is divided into several sub-problems, secondly SI optimization algorithm is used for QP sub-problems optimization, and lastly eigenvectors obtained through Kernel Principal Component Analysis (KPCA) will be input to the SVM classifier to classify the test sample images. Experiments show the validity of proposed algorithm based on SI and SVM which has lower sample training time in target recognition.(4) A high-speed online foreign fiber detection system with machine vision is proposed in this paper. In order to overcome the main bottleneck, namely image processing speed, in machine vision system, a high-speed image acquisition and processing card based on FPGA and DSP chips is developed. The SI optimization algorithm is introduced into the process of foreign fiber detection. The experimental results show that the proposed foreign fiber detection system has more stability and higher detection accuracy.Finally, the main contributions in this work are summarized and further research considerations are put forward also.