工业现场检测设备仪表智能识别与采集方法OA

A data recognition and collection method based on computer vision is proposed aiming at solving the problems of low efficiency and prone to errors in manual collection of traditional industrial instruments.Firstly,the improved YOLOv8 object detection network is used for instrument display recognition.Then,the clustering method based on preset rules is used to cluster the character recognition results and process the data.Finally,the instrument reading recognition results are output.To improve the multi-scale detection performance of the model,the Res2Net(Residual Resolution Network with Squeeze and Excitation,SE-Res2Net)fused with SE is used to replace the residual network in the backbone of the YOLOv8 algorithm.The TPE(Tree-structured Parzen Estimator)is utilized to search for the optimal hyperparameters,and then the model is trained using the instrument image dataset.The average mAP@50 of the object detection model across all categories on the test set reaches 0.906,and the mAP@50 of most categories exceeds 0.9.A regular clustering algorithm is constructed by combining numerical features to handle 12 types of recognition targets.Finally,two actual tests,namely high and low voltage harness detection and simulated square wave voltage output,are conducted to verify the recognition performance of this method for the displayed readings of the measuring equipment.The test results show that the recognition accuracy of this method exceeds 95%,and the average tracking delay is approximately 21 ms,which can meet the data acquisition requirements of industrial field measuring equipment.