Haiqing He's research while affiliated with East China University of Technology and other places

... Moreover, Ziya Ata Yazıcı et al. [16] introduced GLIMS, an attention-guided lightweight multi-scale hybrid network for volumetric semantic segmentation, significantly improving 3D medical image analysis. Yufeng Z et al. [17] proposed a lightweight deep convolutional network with inverted residuals to effectively match optical and SAR images, enhancing the robustness and accuracy of image matching tasks. These studies highlight the potential and effectiveness of lightweight deep learning networks across various fields. ...

... In parallel, wireless sensor networks (WSNs) have been used to monitor the structural health of homes in areas at risk of ground movement. These technologies, which use artificial intelligence and the Internet of Things, represent the vanguard in remote monitoring, contributing to the prevention of harm and the safety of people [37,38]. In geoscience, machine learning methodologies outlined by Dramsch et al. (2020) are primarily categorized into developing alternative models to optimize computational efficiency, crafting models to supplement or replace human intervention, enabling previously unattainable geoscientific activities [39]. ...

... It overcomes the limitations of traditional CNN methods in modeling global dependencies and preserving spatial details, achieving state-of-the-art performance. In the landslide identification task, Huang et al. [46] improved the Swin transformer by incorporating morphological edge analysis to address issues with landslide boundary discretization and irregularity, achieving more accurate landslide boundary extraction in the LuDing area of China. Lu et al. [47] proposed ShapeFormer, a shape-enhanced ViT model designed to effectively handle landslides of various sizes and shapes in remote sensing imagery, enhancing the accuracy of landslide detection. ...

... They integrated dilated convolutional layers in parallel with transformer blocks to improve the network's capacity for extracting features across multiple scales. He et al. (2024) leveraged the CNN-based network (improved EfficicentNet-V2) and transformer-based network (CSwin transformer) to capture local and global semantic information for citrus tree canopy segmentation from 3D data obtained from UAV. Their approach showed superior performance than using some CNN only and some transformer-based only. ...

... Nonetheless, challenges stemming from > REPLACE THIS LINE WITH YOUR MANUSCRIPT ID NUMBER (DOUBLE-CLICK HERE TO EDIT) < the substantial cost and inconvenient installation procedures of these terrestrial sensors have impeded their broad adoption. In contrast, low-altitude photogrammetry based on unmanned aerial vehicle (UAV) developed in the past decade or two has the advantages of low cost, portability, and high accuracy [22], making it an ideal technology for landslide monitoring. In recent years, optical cameras and laser sensors carried by UAVs are often used to obtain dense point clouds to characterize the 3D morphology of landslide surfaces [23], [24], [25], [26], [27]. ...

... To further elaborate the enhancement effect of our proposed method on the visual measurement system, compared to other commonly used coordinate transformation model solution methods such as Quaternion [28] and ICP [29], comparative experiments were conducted. Under the same conditions, calculating coordinate transformation model parameters and visual motion measurement errors. ...

... Since the full convolutional neural network (FCN) [32] realizes the pixel-by-pixel semantic segmentation of images, many classical semantic segmentation models based on convolutional neural networks have been developed, such as UNet [33], Deeplabv3+ [34], FPN [35], PSPNet [36], DANet [37], UPerNet [38], and CCNet [39], extensively embraced in remote sensing semantic segmentation applications [40][41][42][43]. Remote sensing semantic segmentation models dedicated to landcover classification chiefly fall into distinct categories, including the pixel-based CNN, object-based CNN, graph-based CNN, siamese CNN, and ensembled CNN, with the ensembled CNN model emerging as adept at efficiently addressing complex landcover scenarios [44]. ...

... Previous studies have focused on improving the landslide detection process by optimizing the method structure and training data to solve such problems. The optimization of the method structure mainly includes adding residual modules to the deep learning models (Qi et al., 2020;Liu et al., 2020), constructing end-to-end deep learning networks (Yi and Zhang, 2020;He et al., 2022), modifying the feature extraction layer of deep learning models (Liu et al., 2021), and introducing an attention mechanism for landslide detection (Ji et al., 2020;Cheng L B et al., 2021). While optimizing the model structure can improve the accuracy of landslide detection to a certain extent, it may not always be suitable for deep learning models with different structures in practice. ...

... The use of images captured by Unmanned Aerial Vehicles (UAVs) is becoming increasingly popular and widely adopted. These images can also be effectively utilized for texture mapping in CityGML-based 3D city models (He et al., 2022). CityJSON based 3D city models are gaining popularity; however, the literature currently lacks automated solution for texture mapping CityJSON-based 3D city models. ...

... There are a lot of literatures on TLS, e.g. Aydin et al. (2018), Fang (2015), Lv and Sui (2020), Ma et al. (2020), Mahboub (2016), Mercan et al. (2018), Mihajlović and Cvijetinović (2017), Qin et al. (2020), Schaffrin et al. (2012); Uygur et al. (2020); Zeng et al. (2020); Zeng et al. (2022a); Xu et al. (2012); Xu et al. (2023). ...