Adaptable and Efficient Digit Recognition System for Challenging Datasets: A Case Study on Pump Flowmeter Digits

Machine digit recognition from various multi-digit displays is a complex task due to the sheer number of unique digit forms, each varying significantly in shape, size, and orientation. Traditional digit recognition libraries may not perform well for all cases, especially when dealing with digital screens that can be highly variable in terms of style, fonts, colour, contrast, intensity, pixel resolution, digit aspect ratio, and spacing. To address these challenges, we present a digit recognition algorithm that is designed to be fast, easy to use, and highly adaptable. Unlike a single fit-all solution, our system can be easily modified to fit different use cases and applications, incorporating additional layers of flexibility and adaptability. This is desirable since different types of displayed digits may have unique features or characteristics that traditional digit recognition libraries do not capture well. To further demonstrate the efficacy of the proposed system, we tested it on a unique pump-flowmeter digits format, which poses significant challenges for digit recognition algorithms due to the complicated shape and layout of the digits. This paper provides a detailed step-by-step account of our system's development and its performance on this challenging dataset. The presented system achieved an accuracy of 80% on test data, is simple and can be used by researchers, developers, and practitioners working in fields such as handwriting recognition, computer vision, machine learning, image processing, pattern recognition, and neural networks.