In this paper, we study the problem of monocular 3D human pose estimation based on deep learning. Due to single view limitations, the monocular human pose estimation cannot avoid the inherent occlusion problem. The common methods use the multi-view based 3D pose estimation method to solve this problem. However, single-view images cannot be used directly in multi-view methods, which greatly limits practical applications. To address the above-mentioned issues, we propose a novel end-to-end 3D pose estimation network for monocular 3D human pose estimation. First, we propose a multi-view pose generator to predict multi-view 2D poses from the 2D poses in a single view. Secondly, we propose a simple but effective data augmentation method for generating multi-view 2D pose annotations, on account of the existing datasets (e.g., Human3.6M, etc.) not containing a large number of 2D pose annotations in different views. Thirdly, we employ graph convolutional network to infer a 3D pose from multi-view 2D poses. From experiments conducted on public datasets, the results have verified the effectiveness of our method. Furthermore, the ablation studies show that our method improved the performance of existing 3D pose estimation networks.

The accurate segmentation of significant rice diseases and assessment of the degree of disease damage are the keys to their early diagnosis and intelligent monitoring and are the core of accurate pest control and information management. Deep learning applied to rice disease detection and segmentation can significantly improve the accuracy of disease detection and identification but requires a large number of training samples to determine the optimal parameters of the model. This study proposed a lightweight network based on copy paste and semantic segmentation for accurate disease region segmentation and severity assessment. First, a dataset for rice significant disease segmentation was selected and collated based on 3 open-source datasets, containing 450 sample images belonging to 3 categories of rice leaf bacterial blight, blast and brown spot. Then, to increase the diversity of samples, a data augmentation method, rice leaf disease copy paste (RLDCP), was proposed that expanded the collected disease samples with the concept of copy and paste. The new RSegformer model was then trained by replacing the new backbone network with the lightweight semantic segmentation network Segformer, combining the attention mechanism and changing the upsampling operator, so that the model could better balance local and global information, speed up the training process and reduce the degree of overfitting of the network. The results show that RLDCP could effectively improve the accuracy and generalisation performance of the semantic segmentation model compared with traditional data augmentation methods and could improve the MIoU of the semantic segmentation model by about 5% with a dataset only twice the size. RSegformer can achieve an 85.38% MIoU at a model size of 14.36 M. The method proposed in this paper can quickly, easily and accurately identify disease occurrence areas, their species and the degree of disease damage, providing a reference for timely and effective rice disease control.

Agricultural land extraction is an essential technical means to promote sustainable agricultural development and modernization research. Existing supervised algorithms rely on many finely annotated remote-sensing images, which is both time-consuming and expensive. One way to reduce the annotation cost approach is to migrate models trained on existing annotated data (source domain) to unannotated data (target domain). However, model generalization capability is often unsatisfactory due to the limit of the domain gap. In this work, we use an unsupervised adversarial domain adaptation method to train a neural network to close the gap between the source and target domains for unsupervised agricultural land extraction. The overall approach consists of two phases: inter-domain and intra-domain adaptation. In the inter-domain adaptation, we use a generative adversarial network (GAN) to reduce the inter-domain gap between the source domain (labeled dataset) and the target domain (unlabeled dataset). The transformer with robust long-range dependency modeling acts as the backbone of the generator. In addition, the multi-scale feature fusion (MSFF) module is designed in the generator to accommodate remote sensing datasets with different spatial resolutions. Further, we use an entropy-based approach to divide the target domain. The target domain is divided into two subdomains, easy split images and hard split images. By training against each other between the two subdomains, we reduce the intra-domain gap. Experiments results on the “DeepGlobe → LoveDA”, “GID → LoveDA” and “DeepGlobe → GID” unsupervised agricultural land extraction tasks demonstrate the effectiveness of our method and its superiority to other unsupervised domain adaptation techniques.

In recent years, convolutional neural networks (CNNs) have been widely used in hyperspectral image (HSI) classification. However, feature extraction on hyperspectral data still faces numerous challenges. Existing methods cannot extract spatial and spectral-channel contextual information in a targeted manner. In this paper, we propose an encoder–decoder network that fuses spatial attention and spectral-channel attention for HSI classification from three public HSI datasets to tackle these issues. In terms of feature information fusion, a multi-source attention mechanism including spatial and spectral-channel attention is proposed to encode the spatial and spectral multi-channels contextual information. Moreover, three fusion strategies are proposed to effectively utilize spatial and spectral-channel attention. They are direct aggregation, aggregation on feature space, and Hadamard product. In terms of network development, an encoder–decoder framework is employed for hyperspectral image classification. The encoder is a hierarchical transformer pipeline that can extract long-range context information. Both shallow local features and rich global semantic information are encoded through hierarchical feature expressions. The decoder consists of suitable upsampling, skip connection, and convolution blocks, which fuse multi-scale features efficiently. Compared with other state-of-the-art methods, our approach has greater performance in hyperspectral image classification.

Phenotypic analysis has always played an important role in breeding research. At present, wheat phenotypic analysis research mostly relies on high-precision instruments, which make the cost higher. Thanks to the development of 3D reconstruction technology, the reconstructed wheat 3D model can also be used for phenotypic analysis. In this paper, a method is proposed to reconstruct wheat 3D model based on semantic information. The method can generate the corresponding 3D point cloud model of wheat according to the semantic description. First, an object detection algorithm is used to detect the characteristics of some wheat phenotypes during the growth process. Second, the growth environment information and some phenotypic features of wheat are combined into semantic information. Third, text-to-image algorithm is used to generate the 2D image of wheat. Finally, the wheat in the 2D image is transformed into an abstract 3D point cloud and obtained a higher precision point cloud model using a deep learning algorithm. Extensive experiments indicate that the method reconstructs 3D models and has a heuristic effect on phenotypic analysis and breeding research by deep learning.

Semantic segmentation of Polarimetric SAR (PolSAR) images is an important research topic in remote sensing. Many deep neural network-based semantic segmentation methods have been applied to PolSAR image segmentation tasks. However, a lack of effective means to deal with the similarity of object features and speckle noise in PolSAR images exists. Thisstudy aims to improve the discriminative capability of neural networks for various intensities of backscattering coefficients while reducing the effects of noise in PolSAR semantic segmentation tasks. Firstly, we propose pre-processing methods for PolSAR image data, which consist of the fusion of multi-source data and false color mapping. Then, we propose a Multi-axis Sequence Attention Segmentation Network (MASA-SegNet) for semantic segmentation of PolSAR data, which is an encoder–decoder framework. Specifically, within the encoder, a feature extractor is designed and implemented by stacking Multi-axis Sequence Attention blocks to efficiently extract PolSAR features at multiple scales while mitigating inter-class similarities and intra-class differences from speckle noise. Moreover, the process of serialized residual connection design enables the propagation of spatial information throughout the network, thereby improving the overall spatial awareness of MASA-SegNet. Within the decoder, it is used to accomplish the semantic segmentation task. The superiority of this algorithm for semantic segmentation will be explored through feature visualization. The experiments show that our proposed spatial sequence attention mechanism can effectively extract features and reduce noise interference and is thus able to obtain the best results on two large-scale public datasets (the AIR-POlSAR-Seg and FUSAR-Map datasets).