武汉工程职业技术学院

We introduce DepthCropSeg++, a foundation model for crop segmentation, capable of segmenting different crop species under open in-field environment. Crop segmentation is a fundamental task for modern agriculture, underpinning many downstream tasks such as plant phenotyping, density estimation, and weed control. In foundation-model era, a number of generic large language and vision models have been developed. These models have demonstrated remarkable real-world generalization due to significant model capacity and large-scale datasets. However, current crop segmentation models mostly learn from limited data due to expensive pixel-level labelling cost, often performing well only under specific crop types or controlled environment. In this work, we follow the vein of our previous work DepthCropSeg, an almost unsupervised approach to crop segmentation, to scale up a cross-species and cross-scene crop segmentation dataset, with 28,406 images across 30+ species and 15 environmental conditions. We build upon a state-of-the-art semantic segmentation architecture ViT-Adapter, enhance it with dynamic upsampling for improved. detail awareness, and train it with a two-stage self-training pipeline. To systematically validate model performance, we conduct comprehensive experiments to justify the effectiveness and generalization capabilities across multiple crop datasets. Results demonstrate that DepthCropSeg++ achieves 93.11% mIoU on a comprehensive testing set, outperforming both supervised baselines and general-purpose vision foundation models like Segmentation Anything Model (SAM) by significant margins (+0.36% and +48.57% respectively). The model particularly excels in challenging scenarios including night-time environment (86.90% mIoU), high-density canopies (99.86% mIoU), and unseen crop varieties (90.09% mIoU), indicating a new state of the art for crop segmentation.

Remote sensing multi-modal fusion plays an important role in Earth observation and environmental monitoring and has been widely adopted in image recognition and segmentation tasks. However, it is often difficult to guarantee the integrity and validity of multi-source data in practical applications, which greatly reduces the accuracy of traditional multi-modal methods. For accurate and robust land use and land cover classification with missing modalities, we propose a novel unified multi-modal learning network, termed Intermodal Correlation Modeling (InterMC), according to mutual enhancement between feature extraction and fusion. In particular, InterMC adopts a dual-branch encoding structure to extract the inter-modal shared and modality-specific features. It models the intermodal correlation for efficient feature extraction and enforces correlation constraints to meet the requirements of fusion and classification tasks, which thus enables mutual enhancement between feature extraction and feature fusion. Furthermore, we propose a cross-modal fusion module to effectively integrate the extracted cross-modal shared and modality-specific features, and achieve efficient feature fusion guided by the differences and correlations between the modalities. Finally, experiments and ablation studies demonstrate the effectiveness and robustness of the proposed framework in general multi-modal learning scenarios. The combination of improved accuracy and robustness makes the method particularly valuable for practical applications, where system reliability and adaptability are critical.

The task of object goal navigation (ObjNav) requires the agent to locate the given target object within a complex dynamic scene. To successfully accomplish the task, the agent needs to well understand the scenes, make executable decisions with less steps, avoid collisions, and successfully navigate to the target. As a result, efficient environmental perception and scene graph-inspired path planning is important to successfully accomplish the ObjNav task. In this paper, we present a hierarchical scene graph (HSG) contrastive learning, which consists of (1) a multimodal graph mixer that aligns the visual and textual information using open-vocabulary detector with GLIP. It can be regarded as an "eagle eye" to perceive target-related frontiers and suppress irrelevant information, and (2) a graph constructer that takes observed RGBD images to incrementally build a hierarchical scene graph. It acts as the "brain" that memorizes the common scene layout, (3) an action control contrastive learning that takes the graph contextual relationships as input to predict optimal actions to the target. It is treated as the "limbs" of the agent, coordinating and correcting incorrect movements. On the task of ObjNav, experiments on Gibson, HM3D, MP3D, and ProcTHOR demonstrate that navigation plans from the HSG framework achieve significantly higher success rates than existing map-based method, indicating the feasibility of executing navigation utilizing commonsense knowledge from language models leading efficient semantic exploration. Code is available at https://github.com/luosword/HSG4VN.

As the construction of maritime emergency rescue systems worldwide increases, there are more maritime emergency bases along the Yangtze River, yet the mismatch between the layout of these bases and the demand for maritime emergency rescue has emerged. This paper analyzes the impact of marine transportation accident characteristics on emergency demand, uses the CRITIC assignment method to quantify the weights of accident characteristics, and classifies accident levels via the WRSR method. It introduces the equivalent accident number method's research idea and employs the DBSCAN algorithm to cluster and analyze accident blackspots on the Yangtze River mainline. Based on identified blackspots and multiple siting model characteristics, an emergency base siting model considering accident emergency demand is proposed and solved with LINGO. Taking the accident cases in the Nanjing section of the Yangtze River from 2019 to 2021 as an example, a siting scheme meeting the emergency demand of the Yangtze River trunk line is obtained. Model validation results show that at least three maritime emergency facilities need to be built in the Nanjing section to ensure comprehensive coverage of main navigable waters. As the number of emergency facilities grows, their comprehensive emergency response time decreases and multiple coverage of high-risk waters can be achieved. Decision-makers can formulate emergency facility layout plans based on the proposed siting method.

In this article, a distributed probability density function (PDF)-consensus particle filter (PF) algorithm for a class of network systems with degradation is discussed. In order to consider the interaction between component-level degradation and system state, the degradation modelling framework is developed by combining the stochastic degradation process and the state transition model of the system. And the dynamics of the degradation model are described in terms of a Wiener-based process to emphasize the evolution of degradation over time. To reach consensus on the local posterior PDFs at each node in the sense of relative entropy and reduce the communication burden, the state space of the system is divided by a group of weighted grids. Then, the local PDFs are approximated with a combination of the indicator functions and exchange the parameters with neighboring nodes. In order to obtain the particle representation of the fused PDFs, a new importance sampling function is developed to fuse the local grid-based PDFs and to enhance the compatibility of particles with neighboring nodes. A numerical example on target tracking is provided to demonstrate the effectiveness of the proposed scheme.