2results about How to "Good segmentation result" patented technology

The application provides a noise label dynamic correction semi-supervised segmentation framework fusing confidence learning, which can efficiently utilize a small amount of high-quality labeled data and a large amount of low-quality noise data to realize accurate and robust three-dimensional medical image segmentation. In the early training stage, the framework mainly uses high-quality data for full supervision learning to guide the model to learn reliable features and segmentation priori; in the middle and late training stage, low-quality noise labels are gradually introduced, noise region recognition is realized through three-dimensional multi-view slice confidence learning, and an uncertainty-guided dynamic soft correction strategy is adopted to dynamically weight and fuse the teacher model prediction results and the original noise labels, so that the noise labels are effectively utilized while avoiding the interference of false labeling, and finally the segmentation accuracy and generalization ability of the model are improved.

This invention relates to a hyperspectral camouflage target detection method based on spatial-spectral feature fusion. It combines deep learning and spatial-spectral feature extraction to establish a 3D network model suitable for spectral feature learning and target spatial localization. Anomaly detection and spatial-spectral constraints are incorporated into the model to improve detection accuracy. Simultaneously, a false positive / false negative removal training method based on spatial receptive domain is proposed to further adjust the network and improve its training effect. The RX-3DSSRF algorithm effectively compensates for the problems of missed detections and false alarms in the traditional RXD algorithm, achieving good segmentation results on low-altitude hyperspectral camouflage target detection data. By removing false positive features of the target and false negative features of the background, the final optimized target detection result is obtained.