Graph neural networks (GNNs) have gradually become an important research branch in graph learning since 2005, and the most active one is unquestionably graph convolutional neural networks (GCNs). Although convolutional neural networks have successfully learned for images, voices, and texts, over-smoothing remains a significant obstacle for non-grid graphs. In particular, because of the over-smoothing problem, most existing GCNs are only effective below four layers. This work proposes a novel GCN named DII-GCN that originally integrates Dropedge, Initial residual, and Identity mapping methods into traditional GCNs for mitigating over-smoothing. In the first step of the DII-GCN, the Dropedge increases the diversity of learning sample data and slows down the network’s learning speed to improve learning accuracy and reduce over-fitting. The initial residual is embedded into the convolutional learning units under the identity mapping in the second step, which extends the learning path and thus weakens the over-smoothing issue in the learning process. The experimental results show that the proposed DII-GCN achieves the purpose of constructing deep GCNs and obtains better accuracy than existing shallow networks. DII-GCN model has the highest 84.6% accuracy at 128 layers of the Cora dataset, highest 72.5% accuracy at 32 layers of the Citeseer dataset, highest 79.7% accuracy at 32 layers of the Pubmed dataset.

The marine economy has become a new growth point of the national economy, and many countries have started to implement the marine ranch project and made the project a new strategic industry to support vigorously. In fact, with the continuous improvement of people’s living standards, the market demand for precious seafood such as fish, sea cucumbers, and sea urchins increases. Shallow sea aquaculture has extensively promoted the vigorous development of marine fisheries. However, traditional diving monitoring and fishing are not only time consuming but also labor intensive; moreover, the personal injury is significant and the risk factor is high. In recent years, underwater robots’ development has matured and has been applied in other technologies. Marine aquaculture energy and chemical construction is a new opportunity for growth. The detection of marine organisms is an essential part of the intelligent strategy in marine ranch, which requires an underwater robot to detect the marine organism quickly and accurately in the complex ocean environment. This paper proposes a method called YOLOv4-embedding, based on one-stage deep learning arithmetic to detect marine organisms, construct a real-time target detection system for marine organisms, extract the in-depth features, and improve the backbone’s architecture and the neck connection. Compared with other object detection arithmetics, the YOLOv4-embedding object detection arithmetic was better at detection accuracy—with higher detection confidence and higher detection ratio than other one-stage object detection arithmetics, such as EfficientDet-D3. The results show that the suggested method could quickly detect different varieties in marine organisms. Furthermore, compared to the original YOLOv4, the mAP75 of the proposed YOLOv4-embedding improves 2.92% for the marine organism dataset at a real-time speed of 51 FPS on an RTX 3090.

Kinesthetic and tactile information can represent the physical states of objects, encompassing roughness, stiffness, motion, force, and other attributes. The introduction of these can enhance imprecise recognition that relies solely on visual information in cases of light disturbance, occlusion and camouflage. Nevertheless, this task is still challenging due to the heterogeneity among visual, tactile and kinesthetic data. To address this issue, this paper delves into the alignment of heterogeneous data dimensions, the fusion of heterogeneous data features, and the optimization of learning rates for multi-source heterogeneous sensor learning models. Consequently, an effective Visual-Kinesthetic-Tactile Information Fusion (VikitaFusion) network is proposed, which comprises: 1) heterogeneous data extractors that align visual images with tactile and kinesthetic data through image-to-sequence projection; 2) a visual-kinesthetic-tactile Transformer-based domain fusion that mimics human multi-sensory fusion perception through a feature-level fusion block and dynamic fusion blocks; 3) a Periodic Triangulation Learning Rate (PTLR) method aimed at optimizing the learning rate for performance enhancement in multi-source heterogeneous sensor learning models. Extensive experiments demonstrate that VikitaFusion outperforms current state-of-the-art methods with higher recognition accuracy and a lower parameter size. Source code will be made publicly available after the peer review process.