基于多特征图像增强深度卷积神经网络的航天用电子元器件分类算法

为了实现航天用电子元器件的全自动及非接触识别，并减少由照明系统造成的图像亮度不均、偏色等问题对检测结果的影响，通过结合局部、区域和总体三个层次特征提升物体检测精度，提出了一种基于多特征图像增强深度卷积神经网络(MFIE-DCNN)的航天用电子元器件分类算法。MFIE-DCNN算法包含多特征学习和深度学习，其学习过程类似于人类视觉系统，能够对形状、方向和颜色特征进行深度挖掘，突出元器件边界信息，抑制背景杂波干扰。实验结果表明，该算法能够区分电路板板载元器件的种类，检测准确度优于传统算法。对比基于稀疏自动编码器的深度神经网络，检测结果提高了近20%。

Aerospace Electronic Component Classification Algorithm Based on Multi-feature Image Enhanced Deep Convolution Neural Network

In order to implement the automatic and non-contact identification of aerospace electronic component, and to reduce the influence of image brightness unevenness and color-shift caused by the illumination system on the detection results, an aerospace electronic component classification algorithm based on multi-feature image enhanced deep convolution neural network (MFIE-DCNN) is proposed to improve object detection accuracy by combining local, regional, and overall three-level features. The MFIE-DCNN algorithm includes multi-feature learning and deep learning, its learning process is similar to the human visual system. It can deeply excavate the shape, direction and color features, highlight the component boundary information, and suppress background clutter interference. The experiment results show that this algorithm proposed in this paper can distinguish the types of components on the circuit board, and its detection accuracy is better than the traditional algorithm. Compared with the deep neural network based on the sparse autoencoder, the detection result is improved by nearly 20%.