Realization And Optimization Of Image Edge Detection Algorithm For Domestic Accelerators

Image processing has a wide range of applications in remote sensing technology,medicine,security,and industrial production.Image edge detection algorithms are the basis of image processing and analysis for image segmentation,target recognition,and shape extraction.Canny and Sobel algorithms based on gradient calculations are widely used edge detection algorithms today.Foreign mainstream Intel processors,Texas Instruments TIC66 xx signal processors,and NVIDIA GPUs are all adapted to mature edge detection algorithm function libraries.FT-M7002 is a domestic high-performance DSP,adapting the edge detection algorithm to the platform and generating an efficient function library is an urgent problem to be solved.In this paper,Canny and Sobel edge detection algorithms are implemented on the Feiteng platform,and deep optimization is carried out for the architecture and hardware characteristics,providing an efficient edge detection algorithm library for FT-M7002.The main work and contributions of this paper are as follows:(1)Study the principles and characteristics of the Canny and Sobel edge detection algorithms,and implement the code for the FT-M7002 platform,and the test verifies the correctness of the algorithm.(2)According to the hardware characteristics of FT-M7002,a Canny parallel gradient algorithm is proposed,and the three convolution kernels of 3*3,5*5 and 7*7are implemented respectively;the design implements a method based on Pascal's triangulation The 5*5 and 7*7 Sobel extended convolution templates,and the three convolution kernels are optimized in parallel.(3)A parallelized data type conversion interface is proposed,which realizes the algorithm's data type conversion between character type and integer type,character type and floating point type,short integer type and integer type,and improves the parallel algorithm of Canny and Sobel.Scalability and practicality.(4)In-depth study of FT-M7002 compilation tool chain,high-speed cache mechanism,DMA data transmission mechanism,vector storage space and other underlying software and hardware characteristics,through compilation optimization,memory access optimization,loop optimization,software pipeline optimization,double buffering optimization methods,The problem of discontinuous data access is solved,and the algorithm is further optimized from the perspective of the instruction processing time gap and the data transmission gap,which effectively improves the execution efficiency of the algorithm on the target platform.The work done in this article has been implemented and applied on the FT-M7002 platform.The test results show that the performance of the Canny edge detection algorithm on the Feiteng single core is improved by 1.4 to 2.1 times,and the performance of the Sobel edge detection algorithm is improved by 1.075 to 1.525 times.