A Construction Method of Multi-task Classification Network Based on Orthogonal Loss Function

Abstract

The invention provides a method for constructing a multi-task classification network based on an orthogonal loss function. The constructed multi-task classification network simulates the human learning process, and uses a deep convolutional neural network as a hidden layer to simulate the human brain for deep feature extraction. Use a tree classifier as a task-dependent output layer for progressive classification, composing the recognition process into distinct learning tasks. The present invention makes the features obtained by different tasks more in line with their respective needs, and makes the deep features of the same coarse class more aggregated when the classifier completes the coarse classification task, while the deep features of different fine classes are more discrete when the fine classification task is completed, The task output layer features of different classification tasks are distinguished, so that classifiers at different levels can better match the features of different classification tasks, and useless features are removed, thereby improving classification accuracy.

Description

Technical field [0001] The present invention relates to image classification, and in particular relates to a multi-tasking method of constructing classification. Background technique [0002] In recent years, more and more pictures identify the images are widely used in the field of exploration and daily life. This is thanks to the development of deep learning. Current extraction performance is the best tool in depth features convolution neural network. As you know, the depth of convolution neural network not only in the shallow edge information extraction, and semantic information as the number of layers deeper features will become increasingly abstract, the resulting characteristics are also closer to the cognitive human behavior. [0003] Subsequently, multitasking classified network is gradually coming into people's attention. Mutual assistance among different tasks multitasking classified network, a network with different tasks at the same time training, and each has differe...

AI Technical Summary

Problems solved by technology

However, due to the sharing of hidden layer parameters, the extracted features are intersected. Therefore, the multi-task classification network lacks the distinction between the extracted common features and unique features in the process of performing different classification tasks.

Images