Research On Surface Defect Classification Method Of Strip Steel Based On Improved Sparrow Search Algorithm

Strip steel is widely used in fields such as national defense industry,automotive production,aerospace,and light industry.The surface defects generated during the production process of strip steel not only affect the appearance performance,but also directly affect the subsequent processing quality and usage safety.To prevent defective products from flowing into downstream factories,the research on the surface defect detection system of strip steel based on machine vision technology is of great significance for improving the automation production and quality control level of strip steel.In order to deepen the research of machine vision technology for strip surface defect detection systems,this paper proposes an improved sparrow search algorithm(IMM-SSA)that integrates multiple mechanisms,and uses IMM-SSA to optimize the BP neural network classifier to complete the classification of strip surface defects.The specific work is as follows:(1)Analyzed the shortcomings of sparrow search algorithm,and proposed a sparrow search algorithm(IMM-SSA)that integrates multiple mechanisms to address its shortcomings.First,the Tent chaotic mapping sequence is used to diversify the population and the Reverse learning method of dimensional pinhole imaging is used to make the initial population elite,improve the quality of the initial population and improve the global search ability of the algorithm;Secondly,an adaptive factor based on the sigmoid function is introduced to improve the security threshold,expand the search ability of sparrows,and balance the global search ability and local development ability of sparrow algorithms;Thirdly,the Gaussian Cauchy mutation mechanism is proposed to enhance the diversity of sparrow populations,enhance the algorithm’s ability to jump out of local optima and global search,and improve the algorithm’s optimization accuracy.(2)Compare algorithm optimization performance.The IMM-SSA was simulated using23 benchmark test functions,and the genetic algorithm,particle swarm optimization,cuckoo search algorithm,gray wolf algorithm,longicorn whisker algorithm,original sparrow search algorithm and IMM-SSA were simultaneously simulated to compare the iterative performance of IMM-SSA.(3)Apply the improved classifier in this dissertation to classify surface defect images of strip steel.Firstly,the LBP algorithm is used to preprocess the image of surface defects on the strip steel and export the feature vectors.Then apply IMM-SSA to optimize the initial connection weights and thresholds of the BP neural network.Experimental results have shown that the optimized BP neural network classifier(IMM-SSA-BP)has good classification ability,with a classification accuracy of 94.33%,an accuracy of 98.11%,an precision of 94.46%,a recall of 94.33%,a specificity of 98.87%,and a F1 score of 94.31%.(4)Compare the classification ability of IMM-SSA-BP with other classifiers.The original BP neural network classifier,genetic algorithm,particle swarm optimization,cuckoo search algorithm,gray wolf algorithm,longicorn whisker algorithm,sparrow search algorithm and BP neural network are combined to form a classifier,and the classification ability of the above seven classifiers is compared with IMM-SSA-BP.Calculate and compare the accuracy,accuracy,sensitivity,specificity,and values of eight classifiers for inclusion,plaque,cracking,pitting,rolling scale,scratches,and six types of defects,further confirming the classification ability of the classifier proposed in this paper.