Trunk image-oriented open set recognition method

Abstract

The invention discloses a trunk image-oriented open set recognition method, which comprises the following steps of (1) designing a CCN model, training the CNN model by adopting a part of training samples, and constructing a feature extractor; (2) for the feature maps extracted by the feature extractor, calculating the similarity between the feature maps, and clustering the feature maps by adoptinga DBSCAN algorithm; (3) designing a Loss function according to the similarity between the feature maps and the clustering result of the feature maps, and optimizing the parameters of the feature extractor and the weight parameters of the similarity function according to the Loss function; and (4) inputting the trunk images to be classified into an optimized feature extractor, calculating the similarity between the outputted feature images by using the optimized similarity function, and finally obtaining a DBSCAN clustering result of the feature images according to the calculated similarity value. According to the open set recognition method, the open set identification of the trunk images of unknown categories can be well realized.

Description

technical field [0001] The invention belongs to the research field of deep learning algorithms and open set recognition in the field of artificial intelligence, and in particular relates to an open set recognition method for tree trunk images. Background technique [0002] For a long time, the open set recognition method has been a research hotspot in the field of deep learning. This method aims to discover unknown samples from data and extract useful feature information from them, and has wide applications in signal recognition and face detection. Jain et al. proposed an open set recognition method based on multi-classification SVM. They used SVM to learn probability decision scores and rejected unknown samples by setting probability thresholds. Bendale et al. proposed a method of using deep neural networks for open set recognition. They introduced the OpenMax layer and used this layer to estimate the probability that the input sample belongs to the unknown class. Ge et a...

AI Technical Summary

Problems solved by technology

However, the number K of clusters in the algorithm often needs to be given artificially, which limits the application range of K-means in actual scenarios

Images