Mining Intelligent Patterns using SVAC for Precision Agriculture and Optimizing Irrigation (Student Abstract)

The ability to leverage the advances in precision agriculture, computer vision, and edge devices can immensely benefit sustainable agriculture yield. Utilizing the available resources to their maximum requires reliable and intelligent real-time insights to optimize and automate the current agriculture infrastructure. In some countries with low internet penetration rates, such systems need offline and extremely efficient edge deployments. We propose a framework that attends to the trifecta of (i) predicting crop water requirements and irrigating the land appropriately, (ii) providing intelligent insights from aerial images and sensor data for crop management that is fully offline, and (iii) effective post-training quantization and model pruning that leverage the lottery ticket hypothesis - an arbitrarily instantiated network containing a subnetwork that when trained independently will perform as well as the original full network, trained for a similar number of cycles for shrinking the machine learning models and improving latency on the edge.