Research On Machine Vision Target Recognition And Location Technology Based On Deep Learning

With the rise of artificial intelligence,big data,industrial Internet and other technologies,the world's manufacturing is developing towards the direction of intelligence,Internet and automation,and China's manufacturing industry is also in a critical period of intelligent transformation.China has put forward the "made in China 2025" plan,which requires intelligent manufacturing as the core to lead the fourth industrial revolution,and the development of intelligent equipment and products,so that key processes such as grasping,assembly and detection in the manufacturing process can be intelligent,and key positions can be replaced by robots.Machine vision identification technology in the intelligent manufacturing rule artifacts grab,testing has played a big role,but often there will be a size in industrial production put shape change,random and keep out of each product,the traditional machine visual identification technology has been unable to meet the requirements,and is suitable for the emerging technologies of deep learning such scale complex shape change of target detection,research and application of great value.In this paper,based on the national key research and development program(intelligent Chinese cuisine cooking technology and equipment research and development,project number: 2018 yfd0400803),Fried chicken wings,for example,through deep learning method to solve the similar existence scale shape change,surface texture is irregular,put random,mutual existence sheltered complex machine vision target identification and positioning of difficult problems.From visual platform construction to image acquisition,from image preprocessing to mainstream algorithm research,from algorithm improvement to system software implementation,this paper makes an in-depth study on the identification and positioning technology of Fried chicken wings.The work and innovations of this paper are as follows:(1)built a visual experiment platform,determined the detection requirements of Fried chicken wings,and selected and designed the relevant hardware.Completed the production of VOC data set of Fried chicken wings,completed the data cleaning of the data set and the design of 10-fold data enhancement strategy according to the requirements,and prepared sufficient training and test data for the identification and positioning research of Fried chicken wings.(2)to study the current deep learning one of the popular stage in detecting algorithm SSD and two-phase Faster-R-CNN algorithm and related theory,and based on these algorithms for the identification of Fried chicken wings contrast experiment,the experimental results show that the precision and speed in the comprehensive assessment on the SSD model is better than another algorithm,but in the actual operation still exist problems such as delay,caton.(3)SSD deep-fried chicken leg chicken wing recognition and positioning algorithm is improved.First,deep separable convolution is introduced to optimize the trunk network,then k-means Anchor frame clustering is carried out,and Anchor frame design is optimized.Finally,multiple single-layer detection schemes of SSD are replaced by multi-stage feature fusion.Experimental results show that the improved network structure not only surpasses the original algorithm in accuracy,but also reaches the level of real-time detection in speed.(4)based on the design of improved SSD algorithm,a pc-terminal identification and positioning system was implemented.ONNX was used to define the model architecture and optimize it.QT and OpenCV were used to design the client side interaction and model reasoning module.