Research On Food Intelligent Recognition Technology Based On Machine Vision

At present,with the advent of the era of intelligent information technology,artificial intelligence and sensor technology has made great progress,the development of the intelligent robot from the traditional industrial robot has moved towards a service robot and education entertainment robots,etc..Due to machine production has a huge advantage compared with the manual operation is of stable production,high efficiency and cost savings,use machine to replace artificial trend became apparent,and machine vision is to help the robot perceive the surrounding environment,to improve the flexibility and adaptability of robot and intelligent level of one of the key factors.Therefore,it is of great significance to use machine vision technology to research an intelligent dish recognition technology in view of the disadvantages such as inequality and inefficiency distribution of dishes by hand.Based on machine vision technology,combined with traditional image recognition and deep learning image recognition,this thesis proposes an intelligent food recognition technology,which can replace human eyes to recognize dishes.The system adopted the improved Canny edge detection algorithm based on wavelet transform to detect the edge contour and locate the center of the plate,which effectively solves the detection and positioning obstacles caused by noise interference and so on,the transformation model from image coordinate system to world coordinate system is established by camera calibration.In view of the complex and changeable characteristics of food images and the weak robustness of artificial design features,the image processing technology of deep convolutional neural network was applied to recognize dishes,some improvements were applied to the YOLOv3 network model,for example,remove the median-scale and small-scale bounding box detection,and use batch renormalization method for data processing when batch size is mini-batch,which improve the speed of model detection and detection accuracy.Combined with the traditional image processing technology,on the basis of the recognition results of the deep network model,a dish area detection algorithm is designed to complete the task ofdishes area detection.In view of the improved model and the improved model,multiple groups of comparative tests were designed respectively,and the impact of the improved model on the detection results was analyzed.The experimental results showed that the improved model detection speed improved 16.1%,reaching 34.51 fps,detection accuracy reached 96.05%,satisfied the system requirements in real-time and accuracy.