Classification method using distributed classification model

A computer-implemented classification method for enhancing classification accuracy of a target image across network computing terminals by processing computation of a classification model in a distributed manner by splitting the computation to a plurality of terminals, the classification method comprising:

• monitoring, by one of the plurality of terminals or a server, resources of the plurality of terminals, wherein the resources of the plurality of terminals include an amount of computation, a memory size and a network usage;
• allocating subclassification models to the plurality of terminals according to a monitoring result, wherein the allocating comprises selecting, for each terminal, a subset of classes whose execution satisfies a resource threshold associated with that terminal, and wherein the subclassification models are artificial neural networks and the allocation of the subclassification models to the plurality of terminals is dynamically updated in response to changes in the monitored resources;
• generating, by the subclassification models, classification data for the target image, wherein the classification data includes confidence values for classes allocated to the subclassification models receiving, by at least one of the plurality of terminals or the server, the classification data for the target image generated through the subclassification models each distributed to the plurality of terminals;
• and
• determining, by one of the plurality of terminals or the server, a final class for the target image using the received classification data even when classification data from at least one of the subclassification models is not received due to a communication failure or terminal failure, wherein the classes allocated for classification to each of the subclassification models, which have been trained in advance, include a subset of a plurality of target classes allocated to a main classification model, and
• wherein the number of classes allocated to each of the subclassification models is less than the number of target classes allocated to the main classification model,
• wherein the classes allocated for classification to each of the subclassification models and target classes are in the same hierarchy, and
• wherein a distributed execution of the subclassification models is based on the monitored terminal resources of the plurality of terminals and enables classification of the target image that exceeds the resources available at any individual terminal of the plurality of terminals.