Cognitive Machine Vision - overview
Cognitive machine vision solution provides reliable and high speed solutions for product verification leading towards inventory, advisory, early warning, and compliance systems. Our framework provides a set of underlying fundamental machine vision and image processing modules categorized as Detect, Count, Measure and Track. These modules can be quickly re-factored for various IoT-Vision applications based on specific industrial settings. The approaches allow for out-of-the-box usable modules that can be fine-tuned for specific applications even with limited available training data.
How can machine vision help industry?
Currently, the Detect modules have been re-factored for surface anomaly detection in agriculture sector. Specifically, the trained models are utilized in the Plant Pathology app to identify plant diseases and deficiencies manifested on leaf surfaces. The Detect module is also used to identify various anomalies in automobile industry resulting from manufacturing automation defects.
The Count modules are specialized deep learning machines that can be instantiated to identify and stock various objects of interest. In on instantiation, the models are used to estimate farm yield by detecting fruit and fruit bunches (currently grapes and pomegranate). Further, a Measure module will be deployed to ascertain the volume and estimated juice extract from the bunches.
The Track modules are part of DeepSky, that analyses video frames obtained from a sky camera to estimate cloud patterns and cloud movement. DeepSky provides solar irradiance estimates that are comparable with specialized sensors at a fraction of the cost. Further, the Track module allows for near-time forecasting of solar irradiance that is being used as control feedback to align reflectors to enhance the yield.