Research And Implementation Of Holes Location Detection Method For Drip Irrigation Pipe Based On Deep Learning And FPGA

Drip irrigation belt is a common agricultural water-saving product.It needs to drill holes and detect hole location information in the process of production and processing.Missing or wrong drilling will seriously affect the quality of drip irrigation belt.Rapid and accurate detection of the collected holes image during the production and processing of drip irrigation belt is one of the key issues of its production,using convolution neural network to realize holes position in the computer image processing is big volume,power consumption higher shortcoming,the convolutional neural network deployment on FPGA,not only has the speed advantage of parallel computing,but also can solve the above problem.Therefore,it is of great significance to realize embedded image processing of drip irrigation with holes location.This paper mainly does the following work:Implementation of convolutional neural network in PC.Firstly,the data set of holes location information on drip irrigation belt was constructed,and then the convolutional neural network model was built on the PC side.After the training,the prediction accuracy reached 97.62%.Finally,the model parameters were extracted and saved.Deployment of convolutional neural network on FPGA.In the convolution acceleration part,the convolutional kernel and pooling kernel are designed by HLS highlevel synthesis tool,and the decomposition,cyclic optimization and interface optimization are carried out to further improve the performance of the algorithm.The image acquisition module uses OV5640 to extract RGB data.The image display module uses VDMA to drive LCD screen to display the processing results.Vivado software was used to build So C system,IP core port connection and PS terminal driver writing in PL module were completed according to the design requirements,and the construction of drip irrigation belt holes detection system was realized.ZYNQ7020 platform was used for experimental verification.The results show that,at the operating frequency of 100 MHz,verified by self-built drip irrigation belt data set,the accuracy of holes location identification is 96.89%,and the system power consumption is2.028 W,which is 80% less than that of GPU.While maintaining the detection function,the design has the advantages of low power consumption and small volume.The method of hole image detection based on FPGA proposed to this paper provides a new technical idea of drip irrigation belt production.