Tire Defect Detection Based on Faster R-CNN

Abstract

The tire defect detection method can help the rehabilitation robot to achieve autonomous positioning function and improve the accuracy of the robot system behavior. Defects such as foreign matter sidewall, foreign matter tread, and sidewall bubbles will appear in the process of tire production, which will directly or indirectly affect the service life of the tire. Therefore, a novel and efficient tire defect detection method was proposed based on Faster R-CNN. At preprocessing stage, the Laplace operator and the homomorphic filter were used to sharpen and enhance the data set, the gray values of the image target and the background were significantly different, which improved the detection accuracy. Moreover, data expansion was used to increase the number of images and improve the robustness of the algorithm. To promote the accuracy of the position detection and identification, the proposed method combined the convolution features of the third layer and the convolution features of the fifth layer in the ZF network (a kind of convolution neural network). Then, the improved ZF network was used to extract deep characteristics as inputs for Faster R-CNN. From the experiment, the proposed faster R-CNN defect detection method can accurately classify and locate the tire X-ray image defects, and the average test recognition rate is up to 95.4%. Moreover, if there are additional types of defects that need to be detected, then a new detection model can be obtained by fine-tuning the network.

MSANet: efficient detection of tire defects in radiographic images

“…In recent decades, object detection has been widely studied in computer vision fields, such as face recognition [36], autonomous driving [37], defect detection [33][34][35], etc, which mainly consists of two major tasks including object localization and object classification, namely regression and classification. Recently, CNNs have become the mainstream of object detection methods, which can be divided into two-stage and one-stage detectors.…”

“…In recent years, deep learning technology has been widely used in many fields, such as agricultural inspection [22], medical image processing [23][24][25] and defect detection [26][27][28][29][30][31][32][33][34][35]. Defect detection techniques based on deep learning have made great progress by virtue of their dramatically increased performance in feature extraction and representation.…”

“…Defect detection techniques based on deep learning have made great progress by virtue of their dramatically increased performance in feature extraction and representation. Deep learning-based tire defect detection methods can be divided into the following three main categories, namely image classification [26][27][28][29], image segmentation [30][31][32] and object detection [33][34][35]. Tada and Sugiura applied convolutional neural network (CNN) to tire inner surface images for classification [26], which showed effectiveness and robustness in classifying objects with large variations in shape, etc.…”

“…Then, a faster region-based Con-vNet is applied to identify the bubble defects in tire shearography images, which can also minimize the false alarm ratio. In [34], Wu et al proposed a tire defect detection method based on an improved Faster R-CNN in which a homomorphic filtering method was applied to sharpen and enhance the data set and avoid the generation of numerous redundant windows. In their method, an improved zeiler & fergus (ZF) network was used to extract and fuse low-level and highlevel features that were fed into a Faster R-CNN for defect detection.…”

Study on the detection of viscose filament defects based on improved YOLOv5

“…These methods have been widely used in many target detection fields and have achieved a lot of scientific results. Ze-Ju Wu [5] introduced online hard case mining (OHEM) algorithm in Faster-RCNN and achieved 95.7% accuracy in tire appearance defect detection. Liyao Zhang [6] used SSD network for fabric defect detection and improved the accuracy of identifying four different fabric defects to 80%.…”