Design Of Material Sorting System Based On OpenMV Pipeline

Sorting is a very important part in the whole process of pipeline operation.The traditional assembly line usually adopts manual sorting mode,but this method has high labor intensity and low sorting efficiency.With the development of machine vision technology,intelligent sorting devices are constantly updated,and machine sorting is mostly used in assembly lines.However,today ’s automatic sorting devices have the disadvantages of low refinement,single sorting function and high cost.Therefore,it is of great practical value to design an automatic sorting device with diverse functions,high sorting accuracy and high cost performance.This article uses the OpenMV embedded machine vision module for image processing,undergoes comprehensive debugging of the software and hardware system,and is connected to the Alibaba Cloud Internet of Things to complete cloud control.A multifunctional material sorting device is designed to meet the system requirements.This device can intelligently sort materials based on color,shape,size,and neural network extracted material features.It is a functional device that can handle various types of sorting tasks.The research content of this article is as follows:(1)The overall scheme of the assembly line sorting device has been designed based on functional requirements.Build an assembly line sorting experimental platform based on the most widely used belt conveyor on the market,and use STC90C52 RC as the core processor of the lower computer to control multiple functional modules such as sorting module,handling module,Wi Fi module,display module,communication module,and positioning module.(2)In response to the issue of relying on a PC in the image processing process,which makes embedded devices less flexible,the OpenMV machine vision module is selected as the upper computer for image acquisition and algorithm processing.Compare and analyze the advantages and disadvantages of different color spaces such as RGB and Lab,and select the Lab color tracking algorithm with a wider color gamut to complete the sorting of color materials.Comparing the detection performance of Canny and Sobel edge detection operators in OpenMV,a gradient detection template of 45 degrees and 135 degrees was added to the Canny operator with better performance.The improved algorithm reduced the number of edge feature points that need to be traversed for circle detection and quadrilateral detection,improving the efficiency of material shape detection.The method of using a monocular camera imaging principle to detect material size in OpenMV at a fixed height has drawbacks such as poor equipment flexibility and large detection errors.This article proposes a solution that incorporates the Apriltag auxiliary benchmark system.On the one hand,it solves the problem of detection errors caused by uneven conveyor belts,and on the other hand,it can achieve material size detection at any height using OpenMV.(3)For specific sorting tasks where material color and shape cannot be distinguished,this system uses MobileNet V2 as a pre trained model on the cloud AI platform Edge Impulse,and then adds a new trainable layer to extract and classify material features within the model.Use the t-SNE method to reduce dimensionality and visualize the features extracted from the model,and visually demonstrate the model’s effectiveness by comparing the feature extraction images before and after training.Quantify the model using the Int8 method,and deploy the quantified model to the OpenMV end to complete material identification.After the completed cloud project is made public,users only need to provide a small dataset to achieve one click operation from model training to OpenMV deployment,providing a feasible solution for complex work tasks of pipeline intelligent sorting devices.(4)In response to the problem of low intelligence of sorting equipment in assembly lines,a solution is proposed to combine sorting devices with Internet of Things technology,using ESP8266 as a Wi Fi module to send MQTT messages to achieve device access to the Internet of Things platform.By combining the remote control function of the Internet of Things platform with the Wi Fi extension board image transmission function of OpenMV,remote monitoring of PC or mobile devices can be achieved,meeting the application requirements of intelligent management of sorting devices in assembly lines.Comprehensive testing was conducted on the prototype,and experimental data showed that under conditions of stable light and appropriate conveyor belt speed,the accuracy of sorting based on the color,shape,size,and trained model characteristics of the material by the device was above 99%.