The segmentation of 3D point clouds of plant organs, such as leaves and stems, helps to monitor plant growth and is a key step in plant growth phenotype analysis. Compared to point cloud segmentation tasks in other fields, plant point cloud segmentation is more challenging due to the interwoven distribution of various parts such as stems, leaves, and flowers. In this paper, we propose a universal point cloud segmentation network PlantEFRSegnet that can be used for multi-species of plants. The proposed PlantEFRSegnet utilizes a newly designed edge point preservation downsampling module to identify and preserve the points at the edges of plant organs during the downsampling process, in order to assist the segmentation network in learning the contours of various plant organs. PlantEFRSegnet performs supervised feature repair on the point cloud features obtained through downsampling to mitigate the impact of feature loss on segmentation performance during feature embedding. The encoder of the segmentation network is composed of four local feature extraction modules. These four modules can not only extract features but also enhance the features corresponding to points with high contributions in local regions based on point attention mechanism. We evaluated the proposed PlantEFRSegnet on a laser-scanned plant point cloud dataset. Compared with the state-of-the-art approaches, the proposed PlantEFRSegnet achieved better segmentation results.
Using home sensors to provide agitation warnings for community-dwelling dementia patients without ongoing clinical supervision can enable their carers to intervene early during the agitation latent period, thereby reducing unnecessary hospital admissions and harmful events. Most existing studies are based on behavioral, sleep, and physiological data from the previous 24 h to predict patients’ agitation events, which fail to fully capture patients’ recent behavioral details. In this study, monitoring data from the previous 8 × 24 h for dementia patients are used to achieve in-depth mining of patients’ living habits. Furthermore, to address the common clinical problem of extreme imbalance between agitation and normal samples (agitation samples often are extremely scarce), we designed a two-stage agitation early warning model based on symmetric autoencoders and a spatio-temporal network, dubbed SAS-Net. In the first stage, we randomly sample 80% of normal samples and employ multiple symmetric autoencoders to perform feature transformation and pseudo-label construction, and then pre-train the spatio-temporal learning network composed of convolutional networks and a gated Multilayer Perceptron. It aims to learn the patient data’s intrinsic structure, underlying patterns, and effective feature extraction methods. In the second stage, we freeze the parameters of the spatio-temporal learning network and use a balanced dataset consisting of the remaining 20% of normal samples and all agitation samples to reconstruct and fine-tune the top fully connected classifier to improve the recognition performance of agitation samples. The two-stage strategy resolves the problem of ineffective training faced by deep learning models on imbalanced datasets. The experimental results demonstrate the effectiveness of the proposed SAS-Net for agitated behavior early warning.
Synthetic aperture radar (SAR) supports all-weather maritime surveillance, but variations in ship scale, orientation, and scattering, together with sea clutter and coastal infrastructure, can cause missed small ships and false alarms. We propose DSSA-RT-DETR, a dynamic scattering-aware and spatially aligned variant of RT-DETR for SAR ship detection. Its scattering-aware dynamic Inception mixer block (SDIMB) combines multiscale square kernels with matched horizontal and vertical strip-convolution branches. The transposed directional supports provide a symmetry-inspired, axis-balanced operator basis, while independent input-dependent weights preserve orientation adaptivity. Local scattering differences condition the branch weights, making the effective operator sample- and position-dependent. LID-AIFI calibrates the global-attention residual according to local neighborhood differences, and SAF-CCFF aligns upsampled semantic features with shallow spatial details before cross-scale fusion. Controlled ablations support the three modifications. DSSA-RT-DETR achieves 97.55% mAP50, 72.65% mAP50:95, and 71.30% APsmall on SSDD, and 90.95% mAP50, 65.83% mAP50:95, and 67.90% APsmall on HRSID, with 13.471 M parameters and 46.5 GFLOPs. The results show improved aggregate and small-target detection after dataset-specific training, while the symmetry interpretation remains a structural inductive bias rather than an independently isolated accuracy contribution.
Sleep staging is the key step in sleep quality assessment and sleep disorder diagnosis. In clinical practice, doctors usually use polysomnography (PSG) to stage sleep. PSG contains physiological signals such as electroencephalography (EEG) and electrooculography (EOG) and is divided into continuous epochs of fixed duration. In order to achieve more efficient and accurate automatic interpretation of PSG, this paper proposes MM-SleepNet, an automatic sleep staging network based on modal re-weighting and Mamba. MM-SleepNet divides each epoch into several overlapping patches and uses a newly designed patch feature extractor to extract the features of each patch, in order to solve the problem of key waveforms in the patch being obscured by non-key waveforms throughout the epoch. Afterward, MM-SleepNet calculates the importance of different signals on patches at the same time point through a dynamic re-weighting module. MM-SleepNet uses a bidirectional Mamba encoder that considers positional encoding to learn contextual information across epochs to adapt to the transition patterns between sleep stages. The experimental results on the Sleep-EDF and ISRUC datasets show that the proposed MM-SleepNet outperforms state-of-the-art methods. In addition, ablation experiments have demonstrated the effectiveness of various modules in MM-SleepNet.