Five Core Technologies
Multimodal Fusion Technology
A vertical multimodal large model reconstructing the underlying logic of recognition.
The innovation adopts an architecture that combines “visual-language large-model pre-training with specialized fine-tuning on water-meter data,” integrating both semantic and visual multimodal capabilities. breaking through the limitations of traditional “match-based general models” to achieves a leap from “character recognition” to “scene understanding.”
Precision identification technology
Zero-error guarantee, breaking through the bottleneck of measurement accuracy.
Extreme performance metrics
The Mean Absolute Percentage Error (MAPE) is as low as 0.0%, with a hit rate of ±1% reaching 100%. The magnitude error rate is 0%, completely eliminating critical issues such as decimal point/multiplier misalignment and erroneous recognition of wheel rollbacks from “9 to 0.” This fully meets the precision requirements for trade settlement.
Precise analysis of the entire dial
Supports full-type recognition for dial displays featuring rotating wheels,pointers, LCD screens, and hybrid designs. It precisely captures the angle of each digit and hand on the dial, overcoming the limitation of traditional technologies that “only recognize digits but not hands.” Moreover, the output format is controllable, free from extraneous non-digit characters or irregular expressions.
Electromechanical synchronization with zero error
Avoids electromechanical conversion errors in traditional pulse water meters through visual dir-reading technology, meeting the requirements of the "Verification Regulation for Potable Cold Water Meters JJG162-2019" and completely eliminating metering disputes and legal compliance risks.
Generalized Adaptation Technology
0-sample cold start, achieving “no table selection required, no training needed”
0-sample rapid deployment
Without the need for tens of thousands of labeled samples or complex template configurations. The algorithm can automatically identify key information such as the dial’s foreground and background, angle, dial type, and decimal places. It can be launched with zero training samples and will automatically improve its stability as more real-world application samples are added.
Seamless compatibility across all phenotypes
Compatible with all types of meter types including mechanical dial, pointer, and hybrid designs covering DN15-40Omm full-caliber water meters. It seamlessly fits meter housings made of various materials, such as copper, stainless steel, and plastic. Whether it’s a new construction project or the renovation of older residential areas, this solution provides a perfect fit, effectively addressing the longstanding challenges of traditional technologies—such as poor compatibility and inability to support new meter models.
Flexible Full-scenario Deployment
Supports one-click switching between public cloud, private cloud, and edge devices; offers dual-end compatibility for terminal and mobile device shooting ,with provides open-source protocols and terminals, Customizable private deployment solutions tailored to specific needs of government and enterprise clients.
Anti-interference technology
Strong robustness, overcoming the challenges of recognizing extreme scenarios.
Extreme scenario breakthrough
1. Maintains>99.99% recognition rate in common extreme water meter environments such as light spots, water mist, dirt, scratches, bubbles, and blurriness , effectively addressing the core pain points of traditional technologies: “weak resistance to aging and poor anti-interference capability.”
Precise Response to Complex Interference
It can effectively handle multiple types of interference, including dial dirt, scratch-induced reflections, backlight issues, image rotation, refractive distortions, rare fonts, abnormal occlusions, and intermediate states of character wheels, enabling recognition as long as the image is distinguishable by the human eye.
Real-time anomaly alert
Automatically flags abnormal images such as blurriness and freezing. Utilizing AI algorithms, it can automatically detect conditions like cracks and deformations on dial glass, replacing manual on-site inspections and improving the timeliness of anomaly handling.
Iterative Evolution Technology
Self-learning and adaptive—getting more refined the more you use it.
Dynamic performance enhancement
Backed by millions and billions of experimental data points, the model can continuously and automatically optimize itself as it accumulates real-world application samples, achieving “the more it’s used, the newer and more precise it becomes,” while maintaining high stability and a consistently high recognition rate over the long term.
Module Reuse and Rapid Upgrades
The camera module supports plug-and-play replacement, can be repeatedly reused during weekly inspection rotations. Only the base meter and battery need to be replaced . Additionally, it supports replacing the disabled smart meter’s remote transmission module without full meter replacement , thereby reducing the cost of technological upgrades.
Multidimensional Technology Expansion
Based on the core model, value-added services such as freezing warnings and water consumption data analysis can be derived. Supports the expansion of remote meter reading and settlement scenarios for large-caliber water meters, with strong technical extension and secondary development capabilities.


