Pointer Defect Detection Based on Transfer Learning and Improved Cascade-RCNN

Abstract

To meet the practical needs of detecting various defects on the pointer surface and solve the difficulty of detecting some defects on the pointer surface, this paper proposes a transfer learning and improved Cascade-RCNN deep neural network (TICNET) algorithm for detecting pointer defects. Firstly, the convolutional layers of ResNet-50 are reconstructed by deformable convolution, which enhances the learning of pointer surface defects by feature extraction network. Furthermore, the problems of missing detection caused by internal differences and weak features are effectively solved. Secondly, the idea of online hard example mining (OHEM) is used to improve the Cascade-RCNN detection network, which achieve accurate classification of defects. Finally, based on the fact that common pointer defect dataset and pointer defect dataset established in this paper have the same low-level visual characteristics. The network is pre-trained on the common defect dataset, and weights are transferred to the defect dataset established in this paper, which reduces the training difficulty caused by too few data. The experimental results show that the proposed method achieves a 0.933 detection rate and a 0.873 mean average precision when the threshold of intersection over union is 0.5, and it realizes high precision detection of pointer surface defects.

Keywords: defect detection; deformable convolution; online hard example mining; pointer; transfer learning.

Figure 1

Examples of pointer surface defects: (a) hot stamping paper folds; (b) hot stamping paper damage; (c) needle leakage; (d) stains; and (e) hair-like defects.

Figure 2

(a) Hair-like defects have different appearances; (b) stains have different appearances; (c) weak feature of stains; (d) weak feature of hair-like defects; (e) weak feature of hot stamping paper folds; (f) similarity exists in defects of hair-like defects and hot stamping paper folds; and (g) similarity exists in defects of hot stamping paper folds and hot stamping paper damage.

Figure 3

This is the structure of Cascade-RCNN.

Figure 4

This is an example of deformable convolution calculation process.

Figure 5

The reconstruction of the ResNet-50.

Figure 6

Cascade-RCNN’s detection network is improved by online hard example mining.

Figure 7

There are four kinds of defects in the common pointer surface defect dataset: (a) stain; (b) bright spot; (c) filament; and (d) edge gap.

Figure 8

The common pointer defect dataset and the pointer defect dataset constructed in this paper have the same low-level visual characteristics.

Figure 9

This is the process of transfer learning.

Figure 10

The network structure of TICNET.

Figure 11

This is the overall detection framework. The role of the industrial computer is to receive the real-time image of the camera and detect defects, while the role of the PLC is to send detection signals.

Figure 12

The results of defect detection: (a) hot stamping paper folds; (b) hot stamping paper damage; (c) needle leakage; (d) stains; and (e) hair-like defects.

Figure 13

The detection results of four schemes for different situations: (a) hair-like defects have different appearances; (b) stains have different appearances; (c) weak feature of stains; (d) weak feature of hair-like defects; (e) weak feature of hot stamping paper folds; (f) similarity exists in defects of hair-like defects and hot stamping paper folds; and (g) similarity exists in defects of hot stamping paper folds and hot stamping paper damage.

Figure 13

The detection results of four schemes for different situations: (a) hair-like defects have different appearances; (b) stains have different appearances; (c) weak feature of stains; (d) weak feature of hair-like defects; (e) weak feature of hot stamping paper folds; (f) similarity exists in defects of hair-like defects and hot stamping paper folds; and (g) similarity exists in defects of hot stamping paper folds and hot stamping paper damage.