Water level measurement is one of the key issues of hydrological observation.To solve the problems of exist-ing water gauge recognition methods which are poor positioning accuracy and recognition robustness in harsh environ-ments,this paper proposes a multi-feature joint localization-based water gauge recognition method.Firstly,the geo-metric structure and color features of the water gauge are used to roughly locate the water gauge image to obtain candi-date areas.Then,the directional gradient histogram features of the candidate areas are extracted and input into the support vector machine to obtain accurate water gauge positioning;Then,combining morphological operators and pro-jection methods to achieve character and water gauge scale segmentation.Finally,the convolutional neural network LeNet5 is used to recognize normalized and standardized characters and output the water gauge recognition results.To improve the accuracy of water gauge recognition under different perspectives and line of sight conditions,a self built water gauge character library is used to enhance the generalization performance of the character model.Simulation re-sults show that the proposed algorithm,which achieves the recognition accuracy of characters and scales about 99.4％,effectively improves the accuracy and robustness of water gauge recognition in harsh environments.
关键词
水尺识别/联合定位/支持向量机/形态学算子/卷积神经网络LeNet-5
Key words
water gauge recognition/joint localization/support vector machine/convolutional neural network LeNet-5
维普
万方数据