Vision Group Training Day5 - machine learning, image recognition project

7.22 There was no hard task that day , Continue to learn machine learning , Let me introduce the project I'm working on . Now I have three projects in hand, but most of them are similar , One is FPGA Image enhancement of , Mainly for defogging ; A robot arm for visual guidance ; One is FPGA Neural network construction .

A brief introduction ：

One 、FPGA Image enhancement of

This is the whole system architecture , Including data acquisition module ,FIFO Cache module , Filter module , Background update module , Binary module , Morphological processing module , Feature extraction module , Discrimination module . The whole system is powered by 、 The core FPGA chip EP3C25F256、VGA Interface electricity RAM chip JTAG and AS modular 、 Switch output and image sensor interface modules , The schematic diagram adopts OrCAD Design ,PCB use PADS The software design is completed .

This has little to do with machine learning, so I won't go into details . 

Two 、 Vision guided manipulator

Because it is the task given by the tutor , I won't introduce you , It's using jetson Intel D435i, The manipulator is a customized manipulator , The main algorithms are yolo.

3、 ... and 、FPGA Construction of neural network

This is my big innovation project . Don't talk too much, just put the picture , The idea is based on yolov2 Build convolutional neural network , be based on ZYNQ Development .

  The figure above includes the overall architecture 、 Double signed multiplier package correction circuit 、CONV IP Circuit structure 、 Basic pooling unit , The project is still in progress , After that, we will share the overall thinking process .

All in all , machine learning 、 The basic idea of neural network software and hardware is the same , Learning through the principles will be very useful in the process of the project .