Systems and methods for automated inferencing of changes in spatio-temporal images

Abstract

The present disclosure addresses the technical problem of enabling automated inferencing of changes in spatio-temporal images by leveraging the high level robust features extracted from a Convolutional Neural Network (CNN) trained on varied contexts instead of data dependent feature methods. Unsupervised clustering on the high level features eliminates the cumbersome requirement of labeling the images. Since models are not trained on any specific context, any image may be accepted. Real time inferencing is enabled by a certain combination of unsupervised clustering and supervised classification. A cloud-edge topology ensures real time inferencing even when connectivity is not available by ensuring updated classification models are deployed on the edge. Creating a knowledge ontology based on adaptive learning enables inferencing of an incoming image with varying levels of precision. Precision farming may be an application of the present disclosure.