# Color Ring Resistance Classification

Illustration of Color Code Matrix

### Preface

In the field of automatic optical inspection (AOI), printed circuit board (PCB) defect detection is one of the inevitable procedures. Especially, as color ring resistances are in high-frequency occurence in the PCB, the false and missing detection of color ring resistances is particularly considerable.

Recently, some methods directly compare the pixel values of the color rings on the testing resistance with that on the standard resistance. However, the performance may be effected with illumination variance or resistance dispalcement (which means the image only contains a part of the resistance).

### Approach

In order to solve the above problems, I propose a color ring resistance detection network, called RingGCN, based on graph convolutional networks (GCNs). Aiming at the color ring characteristics (i.e., electronic color code) of the color ring resistance, RingGCN is designed in two stages:

• Firstly, extracting the color ring features through a CNN backbone from the input image.
• Then feeding the features with a statistics matrix into a GCN to regress a color code matrix.

As shown in the above figure, the color code matrix $M\in\mathbb{R}^{12\times 12}$ represents the relative location among different color rings, where $12$ derives from the characteristic that 12 kinds of electronic colors are used to indicate the values or ratings of electronic components.

### Performance

The proposed RingGCN achieves a superior performance with only 1.3% false alarm rate (FAR) and 0% missing alarm rate (MAR). The algorithm has been deployed on the AOI machines with a rapid inference speed. Also, the whole solution owns a patent.