The Research On Coal-fired Flame Image Segmentation Method Based On CUDA

Rotary kiln is a high energy consumption,high emission equipment of using pulverized coal as fuel.In order to improve the efficiency combustion of pulverized coal,the combustion condition diagnosis is more and more popular with the coal-fired flame image feature extraction.It is of great significance for the optimization of coal combustion processes,rotary kiln optimal control and energy saving and emission reduction.The traditional method of control rotary kiln is the original artificial fire.In recent years,it is more and more popular that Coal-fired image processing technology is used to control the temperature of rotary kiln.The main drawback of traditional image processing method is the low efficiency,long processing time.The traditional processing techniques and methods are difficult to meet this real-time requirement.CUDA(Compute Unified Device Architecture)is a software development environment and general purpose computing systems based on the GPU(Graphic Processing Unit)using C-like language.GPU has obvious advantage in the parallel processing capability and memory bandwidth than CPU.In this paper,the traditional image processing algorithm is parallel implemented based on CUDA platform.The experimental results show that there is a considerable improvement of operation speed through parallel processing algorithms.The main contents are outlined as following:(1)Through the research on the structure of CUDA programming framework and software system.We can optimize the parallel algorithm of image processing in the tasks division and reasonable allocation of memory.Firstly,tasks are divided into different steps,and obtain the difference between the implementation of efficiency and computational complexity.Secondly,according to the characteristics of each step of the task,the task is divided into host-side and device-side.In the end,according to the characteristics and frequency of data access,GPU further accelerated the speed of image processing through the rational allocation of texture memory and shared memory.(2)Image preprocessing is of great importance in the image processing process,and directly related to the effect of image segmentation and feature extraction,especially in the coal-fired image processing.In this paper,we do research on image preprocessing problem of this easy to be neglected.Histogram equalization,bilateral filtering,edge detection algorithm are to be achieved based on CUDA platform.(3)The flame shape is complex in rotary kiln.Most studies use the method more complex image region segmentation,and then extract features for the division of the"flame","material" and other key regionals.Image segmentation can greatly reduce the volume of data.Therefore,it is very important to fast and accurate segmentation of the interesting region.The realization process of fuzzy C means clustering with a high parallelism.In accordance with the implementation of model CUDA single instruction multiple data,FCM is suitable for parallel computing.In this paper,the algorithm seeking membership matrix,find the cluster centers,and other steps pixel classification is implemented in parallel.(4)In order to prove the validity of the method.The simulation experiment of histogram equalization?bilateral filtering?edge detection and image segmentation was made using coal-fired image based on CUDA.