Design Of Image Visual Saliency Detectiong System Based On PYNQ

With the increasing use of multimedia in the Internet of Things,massive image data brings tremendous pressure to information uploading on the node side of the Internet of Things and server-side information processing.How to extract valuable information in the image at the node of the Internet of Things becomes a thing.The research direction of hotspots in information processing o f networked sensing layer.Under the above background,the subject has inspired the visual attention selection mechanism of human vision and designed an embedded image visual saliency detection system.This system can extract the saliency information of im age data at the sensing node end,thereby reducing the image's Information redundancy.The main research contents of the paper are as follows:First,based on the in-depth analysis of the design requirements,based on the hardware and software co-design thinking,the design of the overall scheme of visual saliency detection system was completed.In the overall design,taking into account the dual requirements of the system in the realization and acceleration of the visual saliency detection algorithm,the heterogeneous processor platform PYNQ was chosen as the embedded platform of the system.Based on this,for the internal functional requirements of the visual saliency detection system,combined with the analysis of the hardware resources of the PYNQ pl atform,the system functions are reasonably distributed to the various processing units of the platform and the core issues in the system design process are clarified.Among them,the task of assigning the visual saliency detection algorithm to the ARM pro cessor is completed,the algorithm acceleration task and the image input/output logic control task are allotted to the FPGA processor,and the information exchange task among the heterogeneous processors is assigned to the DMA.The controller completes.Secondly,in order to improve the algorithm performance of the saliency detection system,this study studied the visual saliency detection algorithm of the image,and completed the implementation of the ARM-based algorithm.In view of the problem that the classical visual saliency algorithm has a single operator and a low detection accuracy,the topic uses a saliency detection method that combines the boundary connectivity operator with the regional contrast operator.Through simulation verification,this method can effectively improve Detecting the accuracy rate and achieving the expected goal.Based on this,the saliency detection algorithm was implemented in the Python environment.Then,in order to improve the processing performance of the saliency detection system,this topic analyzes the principle and characteristics of each processing part on the basis of the implementation of ARM-based algorithms,and accelerates the design of the IP core.Through the software and hardware co-design of the system algorithm,each part is analyzed in parallel,and the color space conversion part and the super pixel vertex matrix multiplication part are selected to perform FPGA parallel acceleration processing,and the hardware path is designed based on Vivado.Realize the data exchange between ARM and FPGA.Finally,this project builds and configures the testing environment of the saliency detection system and performs performance testing.For the PR curve,MAE index,F-measure index,processing speed and other indicators of the system method and the classical method of comparative experiments.After verification,the detection system achieves the expected performance indicators and meets the design requirements.