1. System composition
The intelligent water level image recognition system mainly includes front-end equipment, transmission network, platform software and display terminal, which adopts two forms of timed capture and autonomous capture image, and uploads water meter pictures regularly or as needed.The front-end equipment mainly includes network high-speed cameras, water rulers, 4G traffic cards;The transmission network is mainly transmitted to the information center through the 4G network;Deploy a set of platform software system in the information center, the main task is to receive and store image data in real time, and at the same time to identify the water scale image at the same time, and upload the identified water level data to the display terminal.
2. How the system works
The image recognition system consists of three parts: image segmentation, image feature extraction, and classifier recognition.Among them, the main role of image segmentation is to divide the image into multiple areas;Image feature extraction is the corresponding feature extraction of images of multiple areas;The recognition of the classifier is based on the appropriate classification of the results extracted from the image features.Based on intelligent cameras, the intelligent water level image recognition system uses machine learning and image processing recognition technology to provide monitoring services for the water level of lakes, rivers and urban rainstorms.The intelligent water level image recognition system mainly adopts two forms: timed capture and autonomous capture of images, and uploads water scale pictures and videos regularly or according to business needs.
3. Image analysis technology implementation process
Through the intelligent camera to register the preset position of the station, through the image of the water level, water drop point of the same name and elevation of the template production, to ensure the reliability of the basic data;According to the template matching algorithm, SIFT feature point extraction, RANSAC random sampling consistency and other algorithms, the original map is registered with the template to solve the problem of the same name point and the water scale position of the image caused by camera shaking;According to the image binarization algorithm, accurately find the water mark of the night image water scale under normal conditions and strong light;According to the background color matching template algorithm and the same name point, accurately find the water level line during the day under normal conditions, strong light, water scale reflection, etc.;Based on the semantic segmentation technology of deeplabv3 model, the deeplabv3 model is continuously trained to improve the recognition accuracy of pictures with large recognition errors.According to the intelligent camera algorithm, control the preset point position, perform multi-meter real-time image capture to analyze the water level, and intelligently select the appropriate water scale to analyze the water level.
