基于SURF特征改进的空调标签缺陷检测算法

打开文本图片集

中图分类号：TP317.4 文献标识码：A DOI：10.7535/hbkd.2025yx03010

An improved air conditioner label defect detection algorithm based on SURF features

ZHOU Huizi1'²，LIU Yuelin³，LIU Qing4，LI Jianwul

（1．School of Computer Science and Technology，Beijing Institute of Technology，Beijing 10o081，China； 2.Big Data Center，Zhuhai Gree Electric Appliances Company Limited，Zhuhai，Guangdong 519o7o，China; 3.Ara Institute of Canterbury International Enginering College（Zhongxin International College of Engineering）， ShenyangJianzhu University，Shenyang，Liaoning1lol68，China; 4.School of Economics and Management，Hebei University of Science and Technology， Shijiazhuang，Hebei O50o18，China）

Abstract：Aiming at thebottleneck that deep learning algorithms are not compatible withdevice detectionand new sample colection，aswellaspoordetectiontimelinessandgeneralizationability，atraditionaltemplatematchingdetectionalgorithm basedonSURFfeatures was proposed.Firstly，SURFalgorithm was usedtoextractfeaturesfrom theimage，andthe product quantization theory was used to construct search tres.The matching points were quickly screned basedon spatial position informationof feature points.Secondly，the homographymatrix and afine transformation matrix wereobtained from the matching points，and the two matrices werecombined to scree the "interior points"forofset calculationand image registration.Finaly，combined withtheideaoflocal defect density measurement，thedefect densitywascalculatedby integratingtheregionalforegroundandbackground weighting method，andthequalificationofthelabelwasdeterminedbythe defectdensity.Atthesametime，forthesceneofsmallcharacterswithfewfeaturesandlocalofset，animproved method wasproposed toavoid misjudgment.The results show that thealgorithm improves the stability and detection acuracyof feature point matching. The accuracy，recall and Fl on the self-built data set are 98.67% ， 97.69% and 98.18% ， respectively，which arebettr thanthemainstream methods.The practicalapplicationonthedevice meets thereal-time requirements.Thealgorithmcaneffectively improve thestabilityoffeature pointsandthe detectionacuracy，meet the detection timeliness of equipment，and provide technical reference for its practicability.

Keywords：image processing；defect detection;SURF characteristics；image registration；defect density

随着工业4.0的浪潮推进，标签的质量已经成为衡量企业生产能力和市场竞争力的关键指标之一。（剩余13003字）