Introduction: Semantic segmentation is effective in dealing with complex environments. However, the most popular semantic segmentation methods are usually based on a single structure, they are inefficient and inaccurate. In this work, we propose a mix structure network called MixSeg, which fully combines the advantages of convolutional neural network, Transformer, and multi-layer perception architectures.Methods: Specifically, MixSeg is an end-to-end semantic segmentation network, consisting of an encoder and a decoder. In the encoder, the Mix Transformer is designed to model globally and inject local bias into the model with less computational cost. The position indexer is developed to dynamically index absolute position information on the feature map. The local optimization module is designed to optimize the segmentation effect of the model on local edges and details. In the decoder, shallow and deep features are fused to output accurate segmentation results.Results: Taking the apple leaf disease segmentation task in the real scene as an example, the segmentation effect of the MixSeg is verified. The experimental results show that MixSeg has the best segmentation effect and the lowest parameters and floating point operations compared with the mainstream semantic segmentation methods on small datasets. On apple alternaria blotch and apple grey spot leaf image datasets, the most lightweight MixSeg-T achieves 98.22%, 98.09% intersection over union for leaf segmentation and 87.40%, 86.20% intersection over union for disease segmentation.Discussion: Thus, the performance of MixSeg demonstrates that it can provide a more efficient and stable method for accurate segmentation of leaves and diseases in complex environments.

Image super-resolution (SR) techniques can improve the spatial resolution of remote sensing images to provide more feature details and information, which is important for a wide range of remote sensing applications, including land use/cover classification (LUCC). Convolutional neural networks (CNNs) have achieved impressive results in the field of image SR, but the inherent localization of convolution limits the performance of CNN-based SR models. Therefore, we propose a new method, namely, the dilated Transformer generative adversarial network (DTGAN) for the SR of multispectral remote sensing images. DTGAN combines the local focus of CNNs with the global perspective of Transformers to better capture both local and global features in remote sensing images. We introduce dilated convolutions into the self-attention computation of Transformers to control the network’s focus on different scales of image features. This enhancement improves the network’s ability to reconstruct details at various scales in the images. SR imagery provides richer surface information and reduces ambiguity for the LUCC task, thereby enhancing the accuracy of LUCC. Our work comprises two main stages: remote sensing image SR and LUCC. In the SR stage, we conducted comprehensive experiments on Landsat-8 (L8) and Sentinel-2 (S2) remote sensing datasets. The results indicate that DTGAN generates super-resolution (SR) images with minimal computation. Additionally, it outperforms other methods in terms of the spectral angle mapper (SAM) and learned perceptual image patch similarity (LPIPS) metrics, as well as visual quality. In the LUCC stage, DTGAN was used to generate SR images of areas outside the training samples, and then the SR imagery was used in the LUCC task. The results indicated a significant improvement in the accuracy of LUCC based on SR imagery compared to low-resolution (LR) LUCC maps. Specifically, there were enhancements of 0.130 in precision, 0.178 in recall, and 0.157 in the F1-score.

With the rapid development of computer vision and machine vision, methods based on deep learning have achieved good results in the field of object detection, identification, and tracking. However, for the detection and identification of rebars in smart construction sites, it is very difficult to perform accurate real-time detection of rebars by using object detection technology on the equipment in the field because of the problems of the dense cross-section between bundled bars, the mutual adhesion of cross-section boundaries, and mutual occlusion between cross-sections. To address the above problems, we propose a multi-scale rebar detection network RebarNet with an embedded attention mechanism based on YOLOv5, combining the K-means++ algorithm, attention mechanism, a newly designed SD_IoU Loss, and multi-scale feature fusion, aiming to solve the problems of missed and false detection in dense small object detection. Due to the problems of scarce rebar cross-section datasets, no publicly available large datasets, and weak rebar cross-sectional features, we constructed a new rebar cross-sectional dataset, used a semi-automatic annotation method to annotate part of the dataset, and then used the data enhancement algorithm to expand the rebar dataset. The experimental results show that the average accuracy (mAP) of our proposed RebarNet network is 97.9%, which is comparable to mainstream target detection algorithms such as Faster R-CNN, SSD, RetinaNet, CenterNet, CornerNet, YOLOv3, YOLOv4, and YOLOv5s. mAP0.5 is improved by 8.1%, 13%, 26.4%, 25.8%, 26.2%, 11.7%, 7.6%, and 9%, respectively. In addition, the frames per second (FPS) transmission reaches 89 frames per second, the model weight is only 17.0 MB. In summary, the proposed RebarNet can effectively reduce missed and false detections in the rebar counting detection task based on real-time detection.

Regional land-use change is the leading cause of ecosystem carbon stock change; it is essential to investigate the response of LUCC to carbon stock to achieve the strategic goal of “double carbon” in a region. This paper proposes a residual network algorithm, the Residual Multi-module Fusion Network (R-MFNet), to address the problems of blurred feature boundary information, low classification accuracy, and high noise, which are often encountered in traditional classification methods. The network algorithm uses an R-ASPP module to expand the receptive field of the feature map to extract sufficient and multi-scale target features; it uses the attention mechanism to assign weights to the multi-scale information of each channel and space. It can fully preserve the remote sensing image features extracted by the convolutional layer through the residual connection. Using this classification network method, the classification of three Landsat-TM/OLI images of Zhengzhou City (the capital of Henan Province) from 2001 to 2020 was realized (the years that the three images were taken are 2001, 2009, and 2020). Compared with SVM, 2D-CNN, and deep residual networks (ResNet), the overall accuracy of the test dataset is increased by 10.07%, 3.96%, and 1.33%, respectively. The classification achieved using this method is closer to the real land surface, and its accuracy is higher than that of the finished product data obtained using the traditional classification method, providing high-precision land-use classification data for the subsequent carbon storage estimation research. Based on the land-use classification data and the carbon density data corrected by meteorological data (temperature and precipitation data), the InVEST model is used to analyze the land-use change and its impact on carbon storage in the region. The results showed that, from 2001 to 2020, the carbon stock in the study area showed a downward trend, with a total decrease of 1.48 × ${10}^7$ t. Over the course of this 19-year period, the farmland area in Zhengzhou decreased by 1101.72 ${\mathrm{km}}^2$, and the built land area increased sharply by 936.16 ${\mathrm{km}}^2$. The area of land transfer accounted for 29.26% of the total area of Zhengzhou City from 2001 to 2009, and 31.20% from 2009 to 2020. The conversion of farmland to built land is the primary type of land transfer and the most important reason for decreasing carbon stock. The research results can provide support, in the form of scientific data, for land-use management decisions and carbon storage function protections in Zhengzhou and other cities around the world undergoing rapid urbanization.

Vision transformers (ViT) have the characteristics of massive training data and complex model, which cannot be directly applied to polarimetric synthetic aperture radar (PolSAR) image classification tasks. Therefore, a mixed convolutional parallel transformer (MCPT) model based on ViT is proposed for fast PolSAR image classification. First of all, a mixed depthwise convolution tokenization is introduced. It replaces the learnable linear projection in the original ViT to obtain patch embeddings. The process of tokenization can reduce computational and parameter complexity and extract features of different receptive fields as input to the encoder. Furthermore, combining the idea of shallow networks with lower latency and easier optimization, a parallel encoder is implemented by pairing the same modules and recombining to form parallel blocks, which can decrease the network depth and computing power requirement. In addition, the original class embedding and position embedding are removed during tokenization, and a global average pooling layer is added after the encoder for category feature extraction. Finally, the experimental results on AIRSAR Flevoland and RADARSAT-2 San Francisco datasets show that the proposed method achieves a significant improvement in training and prediction speed. Meanwhile, the overall accuracy achieved was 97.9% and 96.77%, respectively.

In order to improve the reading efficiency of pointer meter, this paper proposes a reading method based on LinkNet. Firstly, the meter dial area is detected using YOLOv8. Subsequently, the detected images are fed into the improved LinkNet segmentation network. In this network, we replace traditional convolution with partial convolution, which reduces the number of model parameters while ensuring accuracy is not affected. Remove one pair of encoding and decoding modules to further compress the model size. In the feature fusion part of the model, the CBAM (Convolutional Block Attention Module) attention module is added and the direct summing operation is replaced by the AFF (Attention Feature Fusion) module, which enhances the feature extraction capability of the model for the segmented target. In the subsequent rotation correction section, this paper effectively addresses the issue of inaccurate prediction by CNN networks for axisymmetric images within the 0–360° range, by dividing the rotation angle prediction into classification and regression steps. It ensures that the final reading part receives the correct angle of image input, thereby improving the accuracy of the overall reading algorithm. The final experimental results indicate that our proposed reading method has a mean absolute error of 0.20 and a frame rate of 15.