Research On Food Packaging Defect Detection Method Based On Deep Learning

Due to the progress of society and the improvement of living conditions,people's requirements for food quality continue to improve,food packaging has also received more and more attention.In view of the serious problem of food packaging defects,enterprises and manufacturers have paid special attention to it.At present,most food processing factories use two ways to detect defects in product packaging: one is manual detection,but the accuracy of this manual detection method can not stably reach the industrial standard,and accompanied by high labor costs,there may be missed detection phenomenon;The other is algorithmic detection based on machine learning,which requires the characteristics of defects to be defined.However,due to the small size and complex shape of food packaging defects,it is difficult to accurately extract their features.Therefore,this paper adopts the convolutional neural network,which can continuously strengthen its feature learning and expression ability,to detect them.In recent years,convolutional neural networks have been more and more widely used in the industrial field,with high accuracy and practicability.Based on this,this paper applies it to the detection of food defects,so that it can quickly and effectively detect the defects on the food surface.This study is mainly carried out from the following two aspects:(1)Research on food packaging defect detection algorithm based on convolutional neural network.This paper discusses the detection and recognition of food packaging defects in detail,and improves the modules of data enhancement,network model and attention mechanism,so as to design a target detection algorithm suitable for food packaging detection environment.(2)Study the system design based on NVIDIA edge computing platform.Aiming at the disadvantages of cloud computing,a food packaging defect detection system based on edge computing platform was designed.In hardware,Jetson TX1,an edge device,is used as a computing platform.The improved defect detection model is deployed in Jetson TX1 and converted into a more efficient ONNX model to realize real-time detection of the model at the edge computing platform.This paper completed the research on food packaging detection based on deep learning.Finally,the proposed food packaging detection algorithm mAP reached 0.95,and the recognition speed was 59.7 ms.Compared with the baseline model YOLOv5,the accuracy was improved by 7.9%,while the recognition speed was improved by 37%.Design packaging defect detection system to promote the efficiency of food packaging defect detection,the liberation of the labor workers,automation and found the quality problems of food packaging in a timely manner,to avoid consuming resources on the artificial defect detection,can also avoid the sales generated when food quality problems,can bring certain economic benefits for the society.