Abstract - This paper presents a new low-cost wireless SCADA (Supervisory Control and Data Acquisition) System using an image-based monitoring module for automatically determining values on Grafana Dashboard Graphs via a Convolutional Neural Network (CNN). The system captures an image of the dashboard screen at pre-determined intervals, crops the image to extract just the graph area, and submits the graph image to a lightweight CNN to obtain the value of the Current Process Variable from the curve. This value is then compared against a user-defined setpoint with the output reflecting the state of the dashboard either “GOOD” (Green) or “BAD” (Red) based on whether the current value falls within the defined tolerance. This non-invasive monitoring approach does not require a direct data API, but instead provides visually verifiable support for displayed trends, as well as support of legacy systems where programmatic access to process variables may be limited. The results of the experimental tests performed with both synthetic and actual dashboard images indicated very high recognition accuracy (>94%) and near real-time inference speed (on an embedded host), demonstrating that an image-based supervisory strategy can provide an additional safety measure for training labs and small installations due to its use of graphical images as an external means of tracking process values in industrial settings.

Keywords: Convolutional Neural Network; Chart Image Recognition; Dashboard Monitoring; Grafana; SCADA; Image-based Meter Reading; Setpoint Verification; Edge AI; Real-time Monitoring; Industrial Automation; OCR – Optical Character Recognition.

Downloads: | DOI: 10.17148/IJIREEICE.2026.14203