The target detection algorithm can only provide the information of the normal bucket teeth in the image to be detected. Because the labels provided are the same, the detected images can only be labeled in turn. The position of the faulty bucket teeth cannot be directly determined only by the information provided by the target detection algorithm. As shown in the figure, for the detection result of the target detection program on the image of a faulty bucket tooth, the bucket teeth in the figure are given the same label and the number of detected bucket teeth is the same, And the position relationship between the bucket teeth is the same, so it is impossible to determine which bucket tooth has failed.
Through the introduction of the bucket tooth system, it can be seen that each detection is after the bucket is unloaded, so the image at some time can be used as a template (such as initial detection). After obtaining the position relationship between the normal bucket teeth, combined with the predicted frame size and position relationship in the fault image to be detected, the fault position of the bucket tooth can be judged. Considering that if the fault occurs on the bucket teeth at both ends and cannot be determined according to the position relationship, it is also necessary to calibrate the bucket teeth at both ends.
Due to the bad working condition of bucket teeth, that is, the background of bucket is complex and changeable, the image segmentation based on deep learning method can cut the background of the image from the semantic level, and the parameters trained in Chapter 3 can be used for the fine-tuning training of image segmentation network. Secondly, the image after background cutting is easy to be processed into binary image, which makes it easier to compare some properties of bucket teeth (such as the length of bucket teeth), so as to make a preliminary judgment on the wear degree of bucket teeth. Therefore, based on the image segmentation, the contour features will be extracted by using the processed segmentation image, so as to solve the calibration problem of bucket teeth at both ends and realize the preliminary judgment of the wear degree of bucket teeth.