Small sample insect image recognition method based on visual Transform

Abstract

The invention discloses a visual Transform-based small sample insect image recognition method, which comprises the following steps of: firstly, searching images of various insects by utilizing a search engine, and manually labeling the images with labels; then, constructing a pre-training model taking a visual Transform as a core, and carrying out optimization training on parameters in the pre-training model by utilizing the training set; removing a classifier in the pre-training model, randomly extracting a small number of image samples of each type of insects in the training set and the test set, inputting the image samples into a visual Transform to extract image features, and calculating an average value of each type of samples as prototype features of each type of insect images to be stored; and finally, acquiring insect images on line, inputting the insect images into Transformer to extract image features, calculating the distance between the insect images and each type of insect image representation, and outputting the insect type with the closest distance as the type of the image. According to the method, the insects are classified and identified by using a small number of training samples, and the technical problem that a large number of image samples are needed during convolutional neural network training used in current insect identification can be solved.

Description

technical field [0001] The invention relates to a small sample insect image recognition method based on a visual Transformer, and belongs to the field of computer vision. Background technique [0002] Insects are the most numerous animal groups on earth, and they come in many varieties and shapes. From the perspective of human self-interest, these insects can be generally divided into two categories: pests and beneficial insects. Pests will harm the growth of crops and ornamental flowers, and will bring serious economic losses to human beings, while beneficial insects may bring nutritious food or rich industrial materials to human beings. Therefore, the effective identification of insects is of great significance for protecting the ecological environment and promoting the development of social production. Compared with the traditional manual recognition method, the labor cost of the automatic insect recognition method based on computer vision technology will be greatly red...

AI Technical Summary

Problems solved by technology

However, the technical problems that still exist in the above method are as follows: (1) The category of insect recognition is limited to Odonata or Diptera insects, and there is still a lack of image data sets containing four orders of insects; (2) The ResNet 50 used Or convolutional neural networks such as ResNeXt have a large parameter capacity, requiring a large number of image samples for optimization training to ensure good recognition performance

Images