A fine-grained vehicle recognition system and method based on text feature fusion

Abstract

The invention discloses a fine-grained vehicle recognition system and method based on text feature fusion. The system includes: a feature extraction module, a classification layer, a text representation network, a similarity calculation module, a fusion label calculation module, and a divergence loss calculation module; method Including: step S1, constructing a fine-grained vehicle image classification dataset; step S2, extracting features from training images; step S3, classifying image feature vectors; step S4, inputting labels of each subcategory of the dataset into pre-trained text Representing the network; step S5, through the image feature vector and the word vector of the image label; performing weighted fusion of the obtained enhanced label distribution and the original label vector; step S6, using the similarity between the predicted label distribution and the weighted and fused label distribution as a loss, Guide the training of the whole system; step S7, inference stage, perform feature extraction and classification layer on the image to be tested, and determine the image category according to the predicted label distribution.

Description

technical field [0001] The invention relates to the technical field of computer vision recognition, in particular to a fine-grained vehicle recognition system and method based on text feature fusion. Background technique [0002] Fine-grained image classification has a wide range of research needs and application scenarios in both industry and academia. Related research topics mainly include identifying different types of birds, dogs, flowers, cars, and airplanes. In real life, there is a huge application demand for identifying different subcategories. The difference and difficulty of the fine-grained image analysis task compared with the general image task is that the granularity of the category to which the image belongs is finer, the difference between the subcategories is subtle, and the difference within the subcategory is large. Therefore, not only for computers, but also for ordinary people, the difficulty and challenge of fine-grained image tasks is undoubtedly huge...

AI Technical Summary

Problems solved by technology

[0005] Considering that there are a large number of similar categories in fine-grained images, some subtle differences are difficult to distinguish even for humans, and it is easy to introduce noise labels in the labeling process

In the classification process, the use of one-hot encoding as the original label is too arbitrary, ignoring the correlation between labels, making the noise label play a greater negative role in network training, thereby affecting network convergence

Images