Place recognition technology is very important for autono-mous driving. To realize the large-scale recognition task of 3D point clouds, we propose a large-scale 3D point cloud place recognition framework based on graph neural networks, which combines local and global features. In extracting features, instance segmentation is performed on the large scene point clouds first, and then the GNN network trains each segmented instance to obtain local attribute features. We construct a graph model with each object as a node and the relationship between them as edges, then obtain the global topological structure features of the scene. In calculating similar scores, we calculate the similarity vector of the global and local feature through a similarity network and cosine similarity, respectively. Finally, we fuse the similarity vectors and calculate the final similarity score. This paper uses the SemanticKitti and nuScenes datasets to verify the proposed method. Compared with the state-of-the-art deep learning-based place recognition method, the proposed method achieves the best results in the SemanticKitti and nuScenes datasets.

Anomaly detection is widely used in manufacturing and medical imaging. We propose a self-supervised defect detection method based on multi-scale feature fusion, which can effectively improve the detection and localization accuracy. The method of pseudo-defect construction was used to enhance the training data. To make the pseudo-defects more realistic, the extreme point of feature heatmap was used as the anchor point of the defect area, and the defect image was fused with the original image to construct the pseudo-defect. A multi-scale feature fusion network was proposed that utilizes the self-attention mechanism and the interaction between multi-scale features to extract semantic features containing rich contextual information to improve detection and localization accuracy further. The proposed method achieved competitive experimental results on both the MVTec AD and Chest X-ray datasets. Compared with other pseudo-defect simulation methods, the heatmap-based pseudo-defect construction method improves by at least 2%. It achieves comparable results with other state-of-the-art defect detection methods.