基于深度学习的电子元器件快速检测算法研究

针对电子元器件装配过程中由于元器件体积小、外观相似,导致工人在长期高强度的工作下极易误识别、误装配元器件的问题,提出了一种基于深度学习的检测算法ETS-Net实现电子元器件的快速准确检测。算法引入深度可分离卷积,减少模型参数量和运算量,降低了模型的复杂度。提出一个轻量化高性能卷积神经网络提取具有分辨力的抽象特征,采用K-Means聚类并微调得到适合本场景的锚框,使用高效率的区域提议网络获取高质量的预选框。其次利用两个并联的全连接层预测类别并再次调整预选框,采用非极大抑制排除冗余检测结果。实验结果表明,该算法在电子元器件装配机器人视觉检测任务上具有较高的鲁棒性和实时性。

Research on fast electronic component detection algorithm based on Deep Learning

Aiming at the problem that workers are easy to identify and assemble components by mistake under long-term and high-intensity work due to the small volume and similar appearance of components in the assembly process of electronic components, a detection algorithm ETS-Net (Efficient Two-Stage Network) based on deep learning is proposed to realize the rapid and accurate detection of electronic components. The algorithm introduces depthwise separable convolution to reduce the amount of model parameters and computation, and eliminate the complexity of the model. A lightweight and high-performance feature extraction network is proposed to extract discriminative features, K-means clustering and fine-tuning are adopted to obtain a set of anchor boxes suitable for the scene, an efficient regional proposal network is introduced to obtain high-quality proposals. And then, two sibling fully-connected layers are used to predict classes and adjust proposals again, and non-maximum suppression is introduced to reduce redundant detection results. The experimental results show that the algorithm has high robustness and efficiency in the visual detection task of electronic component assembly robot.