Digital Display Recognition towards Connected Sensing Systems for Precision Agriculture

Many sensing systems deployed in the field and factories have limited communication interfaces to transfer the data they generate to a central hub or gateway. Sometimes, it is an active legacy system that has not been upgraded in a long time. While in other cases, the system has limited communication interfaces by design to limit costs. In the precision agriculture value-chain, sensing systems range widely from low-cost frugal COTS handheld devices on the field to those in the crop and commodity processing unit in the factory. A common thread we observe in these systems is the presence of a simple or complex layout of a seven-segment digital display. Accurate digital display recognition is not without its challenges given the range of devices, displays, and other operational constraints where a recognition system to be scalable would be expected to generalize well across all devices while giving reliable outputs for each. We present our results with some typical displays we encounter in the precision agriculture scenarios where performance with available OCR systems is compared with proposed approaches. We propose to transform such scenarios where display information captured as images is reliably converted into measurements used to generate actionable insights in real-time.