Fine-grained image classification method based on depth convolution neural network

Abstract

The invention discloses a fine-grained image classification method based on a depth convolution neural network, comprising the following steps: 1) preparing a fine-grained image classification data set, and dividing the training data into a training data set and a verification data set; 2) building a depth convolution neural network model for fine-grained image classification, training the model with a training data set, and saving the network model parameters when the trained model reaches the set precision on the verification data set; 3) calculating the tags of test image classification orclassification accuracy of test data sets by using the trained model. The fine-grained image classification framework provided by the invention is independent of and can be applied to any DCNN structure, and has good portability.

Description

technical field [0001] The invention belongs to the technical field of computer vision image classification, and in particular relates to a fine-grained image classification method based on a deep convolutional neural network. Background technique [0002] The difference between the fine-grained image classification task and the general image classification task is that the granularity of the category to which the image belongs is finer, and the difference between different fine-grained object classes is only reflected in the subtleties. The main challenge of fine-grained image classification lies in the inter-class similarity and intra-class diversity. On the one hand, the visual differences between different fine-grained classes are only reflected in subtleties; on the other hand, due to the influence of location, viewing angle, lighting and other conditions, even instances of the same class may have large changes. Intra-class visual differences. For example, the differe...

AI Technical Summary

Problems solved by technology

[0003] The part-based algorithm fine-grained image classification method first detects different parts of the target object, and then increases the difference between classes and reduces the difference within the class through local feature modeling. This type of method is highly dependent on accurate component detection and is prone to occlusion. , angle and posture effects

The method of applying contractive and triplet loss functions to train deep convolutional neural networks has slow convergence speed and high computational complexity

Images