Abstract

åŽŸæ²¹ä½œä¸ºä¸€ç§é‡è¦çš„æˆ˜ç•¥ç‰©èµ„ï¼Œåœ¨æˆ‘å›½ç»æµŽå’Œå†›äº‹ç­‰å¤šä¸ªé¢†åŸŸå‡èµ·åˆ°é‡è¦ä½œç”¨ã€‚æœ¬æ–‡æå‡ºä¸€ç§åŸºäºŽæ·±åº¦å­¦ä¹ çš„ç›®æ ‡æ£€æµ‹æ¨¡åž‹TCS-YOLO(Transformer-CBAM-SIoU YOLO),该模型在YOLOv5çš„åŸºç¡€ä¸Šè¿›è¡Œä¼˜åŒ–ï¼ŒåŒæ—¶åŸºäºŽå‰æž—ä¸€å·å ‰å­¦é¥æ„Ÿå«æ˜Ÿå½±åƒæ•°æ®é›†è¿›è¡Œå®žéªŒï¼Œå¯¹å ¨çƒèŒƒå›´å† çš„å‚¨æ²¹ç½è¿›è¡Œè¯†åˆ«ä¸Žåˆ†ç±»ã€‚ä¼˜åŒ–å† å®¹åŒ æ‹¬ï¼šæ·»åŠ åŸºäºŽTransformer架构的C3TR层对网络进行优化;使用CBAM(Convolutional Block Attention Moduleï¼‰åœ¨ç½‘ç»œå±‚ä¸­æ·»åŠ æ³¨æ„åŠ›æœºåˆ¶ï¼›ä½¿ç”¨SIoU(Scale-Sensitive Intersection over Union) loss代替CIoU(Complete Intersection over Union) loss作为定位损失函数。实验结果表明:与YOLOv5相比,TCS-YOLO的模型复杂度(Giga Floating Point of Operations,GFLOPs)平均减少3.13%,模型参数量(Parameters)平均减少0.88%,推理速度(Inference Speed)平均降低0.2 ms,mAP0.5(mean Average Precision)平均提升0.2%,mAP0.5∶0.95平均提升1.26%。与此同时,将TCS-YOLOæ¨¡åž‹ä¸Žé€šç”¨ç›®æ ‡è¯†åˆ«æ¨¡åž‹YOLOv3,YOLOv4,YOLOv5和Swin Transformer进行对比实验,TCS-YOLO均体现出了更高效的特点。TCS-YOLOæ¨¡åž‹å¯¹å ¨çƒå‚¨æ²¹ç½çš„ç›®æ ‡è¯†åˆ«å ·æœ‰é€šç”¨å¯è¡Œæ€§ï¼Œå¯ä¸ºé¥æ„Ÿæ•°æ®åœ¨èƒ½æºæœŸè´§é¢†åŸŸæä¾›æŠ€æœ¯å‚è€ƒã€‚

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call

Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.