卷积神经网络圆检测方法用于多硬币检测

Convolution Neural Network Circle Detection Method for Multi-coin Detection

为了开发一种检测精度高,检测速度快的圆检测方法,研究者们进行了大量的研究.然而,现有的圆检测方法都依赖于边缘检测器获取的边缘图进行计算,边缘图不仅包含大量无效边缘,而且将有效的圆弧边缘也混杂为一体,不利于多圆检测.受到卷积神经网络在其他领域成功的启发,本文提出一种基于卷积神经网络的圆检测方法.本文方法利用目标检测技术和语义分割技术将多圆检测任务划分为多个单圆检测任务,并且能准确地提取圆的边缘信息(不包含背景和纹理的边缘).为了训练检测模型和验证方法的有效性,本文收集了硬币图像进行标注作为数据集,并通过实验对比三种优秀的圆检测方法.实验结果表明,本文的圆检测方法获得了较高的检测精度,在测试集上优于所有对比方法.

In order to develop a circle detection method with high detection accuracy and fast detection speed,researchers have conducted a lot of research.However,the existing circle detection methods all rely on the edge map obtained by the edge detector for calculation.The edge map not only contains a large number of invalid edges,but also mixes effective arc edges into one,which is not conducive to multi-circle detection.Inspired by the success of convolutional neural networks in other fields,this paper proposes a circle detection method based on convolutional neural networks.The method proposed in this paper uses target detection technology and semantic segmentation technology to divide the multi-circle detection task into multiple single-circle detection tasks,and can accurately extract the edge information of the circle(not including the background and texture edges).In order to train the detection model and verify the effectiveness of the method,we collect coin images for annotation as a data set,and compare three excellent circle detection methods through experiments.The experimental results show that the circle detection method proposed in this paper obtains a high detection accuracy,which is better than all comparison methods on the test set.