基于改进yolov网络的外观检测研究

Research on Appearance Detection Based on Improved YOLOv Network

摘要:

外观检测涉及对图像或视频中的物体进行准确和高效的识别和定位,为了解决物体表面小尺寸目标检测的问题,研究通过优化YOLOv3网络模型,引入多尺度检测和深度可分离卷积技术来提高检测精度和模型效率,以增强对小尺寸目标的识别能力,再采用深度可分离卷积技术来减少计算量,并提高模型的训练效果。实验结果表明,改进后的算法在物体表面小尺寸外观检测方面表现出明显的提升,平均精度达到71.52%,比原始模型提高11.37个百分点。同时,通过减少计算量和提高模型速度,实现了35.6帧/秒的检测速度。研究可以优化算法,提高小尺寸目标检测的准确性和鲁棒性,推动其在计算机视觉领域的广泛应用。

Abstract:

Appearance detection involves accurate and efficient recognition and localization of objects in images or videos. To address the issue of detecting small-sized objects on object surfaces, this research improves the YOLOv3 network model by introducing multi-scale detection and depthwise separable convolution techniques to enhance detection accuracy and model efficiency. The deep separable convolution technique is employed to reduce computational complexity and improve model training effectiveness. Experimental results demonstrate a significant improvement in detecting small-sized objects, with an average precision of 71.52%, which is an 11.37% increase compared to the original model. Moreover, the improved algorithm achieves a detection speed of 35.6 frames per second by reducing computational overhead and enhancing model inference speed. This research has the potential to optimize the algorithm and enhance the accuracy and robustness of small-sized object detection, thereby promoting its wide application in the field of computer vision.