Crop diseases pose a severe threat to global agricultural productivity and food security. While deep learning has revolutionized crop disease recognition, prevailing methods are currently constrained by a closed-set assumption, rendering them prone to overconfident errors when encountering unknown diseases in real-world open environments. To address this, we propose the Prototype-Aware Gated Multi-Scale Branch Network (PAG-MSBNet), a novel Open-Set Recognition (OSR) framework tailored to the unique challenges of crop disease imagery. Our approach features a multi-branch architecture that captures complementary fine-grained features through dedicated pathways. Each branch is equipped with a Multi-Scale Spatial-Channel Attention (MSSCA) module to enhance focus on polymorphic lesion regions across scales. Central to our framework is a Prototype-Aware Gating (PAG) mechanism, which dynamically fuses multi-branch predictions by evaluating their semantic alignment with learnable class prototypes, thereby enabling adaptive and robust feature integration. Comprehensive experiments on public benchmark datasets, including Paddy Doctor and PlantVillage, demonstrate that PAG-MSBNet consistently outperforms state-of-the-art OSR methods. It achieves superior performance in both closed-set classification accuracy and the crucial ability to reject unknown categories, while maintaining remarkable robustness across varying degrees of task openness. Our model provides a reliable and effective solution for practical crop disease diagnosis in complex agricultural settings.

Site-specific weed management (SSWM) provides precise weed control and reduces the use of herbicides, which not only reduces the risk of environmental damage but also improves agricultural productivity. Accurate and efficient weed detection is the foundation for SSWM. However, complex field environments and small-target weeds in fields pose challenges for their detection. To address the above limitations, we developed WeedDETR, a real-time end-to-end detection model specifically designed to enhance the detection of small-target weeds in unmanned aerial vehicle (UAV) imagery. WeedDETR incorporates RepCBNet, a backbone network optimized through structural re-parameterization, to improve fine-grained feature extraction and accelerate inference. In addition, the designed feature complement fusion module (FCFM) was used for multi-scale feature fusion to alleviate the problem of small-target weed information being ignored in the deep network. During training, varifocal loss was used to focus on high-quality weed samples. We experimented on a new dataset, GZWeed, which contains weed imagery captured by a UAV. The experimental results demonstrated that WeedDETR achieves 73.9% and 91.8% AP0.5 (average precision at 0.5 intersection over union threshold) in the weed and Chinese cabbage [Brassica rapa subsp. chinensis (L.) Hanelt] categories, respectively, while achieving an inference speed of 76.28 frames per second (FPS). In comparison to YOLOv5-L, YOLOv6-M, and YOLOv8-L, WeedDETR demonstrated superior accuracy and speed, exhibiting 3.5%, 6.3%, and 3.6% higher AP0.5 for weed categories, while FPS was 14.9%, 12.9%, and 1.4% higher, respectively. The innovative architectural design of WeedDETR significantly enhances the detection accuracy of small-target weeds, enabling efficient end-to-end weed detection. The proposed method establishes a solid technological foundation for UAV-based precision weeding systems in field conditions, advancing the development of deep learning-driven intelligent weed management.

Recently, vision transformer (ViT)-based methods have made breakthroughs on plant disease recognition tasks and have surpassed convolutional neural network (CNN)-based methods. They are now considered the state-ofthe-art for such methods. However, ViT-based methods usually encode and decode images through global modeling, which introduces a large amount of noise information when dealing with plant disease images in complex environments. In addition, plant disease images in complex environments have significant intra- and inter-class differences, further limiting the performance of ViT-based methods. To address the above limitations, we propose the discriminative information enhanced contrastive vision transformer, in short DIEC-ViT, for plant disease recognition in complex environments. DIEC-ViT contains two key modules, namely, the discriminative information enhancement (DIE) module and the contrastive learning (CL) module. Specifically, the DIE module enhances the perception of discriminative regions of the ViT and suppresses complex backgrounds by counting multi-head self-attention for multi-levels of class tokens. To cope with the problem of intra- and inter-class differences in plant disease images, the CL module is introduced into the ViT to optimize the feature space by reducing the distance between positive pairs and increasing the distance between negative pairs. Extensive experiments verify the effectiveness of the two modules. In addition, DEIC-ViT outperforms state-of-the-art methods with three field plant disease datasets. The obtained results indicate the potential of our approach to drive further development of ViT in the field of plant disease monitoring.

BACKGROUND: Semantic segmentation of weed and crop images is a key component and prerequisite for automated weed management. For weeds in unmanned aerial vehicle (UAV) images, which are usually characterized by small size and easily confused with crops at early growth stages, existing semantic segmentation models have difficulties to extract sufficiently fine features. This leads to their limited performance in weed and crop segmentation of UAV images. RESULTS: We proposed a fine-grained feature-guided UNet, named FG-UNet, for weed and crop segmentation in UAV images. Specifically, there are two branches in FG-UNet, namely the fine-grained feature branch and the UNet branch. In the fine-grained feature branch, a fine feature-aware (FFA) module was designed to mine fine features in order to enhance the model's ability to segment small objects. In the UNet branch, we used an encoder-decoder structure to realize high-level semantic feature extraction in images. In addition, a contextual feature fusion (CFF) module was designed for the fusion of the fine features and high-level semantic features, thus enhancing the feature discrimination capability of the model. The experimental results showed that our proposed FG-UNet, achieved state-of-the-art performance compared to other semantic segmentation models, with mean intersection over union (MIOU) and mean pixel accuracy (MPA) of 88.06% and 92.37%, respectively. CONCLUSION: The proposed method in this study lays a solid foundation for accurate detection and intelligent management of weeds. It will have a positive impact on the development of smart agriculture. (c) 2024 Society of Chemical Industry.

This study examines credit spreads on municipal bonds through an innovative cross-application approach. Based on the exacerbated economic uncertainty and the unique position of municipal bonds in the Chinese credit market, the study of municipal bonds’ credit spread prediction is necessary. In this study, we propose a hybrid RF-DWT-SAM-GRU model, and we find that the random forest (RF) for feature selection enhances the model performance well, while the discrete wavelet transform (DWT) for time-series decomposition is more suitable for this study’s dataset compared to CEEMDAN (Complete Systematic Empirical Modal Decomposition of Adaptive Noise), and the gated recurrent unit (GRU), which is augmented by the self-attention mechanism (SAM), is more accurate in captures the fluctuations and trends in credit spreads of municipal investment bonds, thus improving the prediction accuracy. By comparing the other models, this paper finds that the hybrid model has higher forecasting accuracy in the prediction of municipal bonds.

Analyzing investor sentiments correctly plays a crucial role in preventing potential market risks. However, in the selection of data, existing studies predominantly focus on textual data presented by investors, while overlooking the supplementary role of other financial data in sentiment analysis. For instance, related background news, as well as responses from company secretaries to investor questions on investor interactive platforms. This oversight results in discrepancies between the analyzed sentiments and the actual sentiments. In this paper, we propose an approach that combines GPT and RoBERTa for investor sentiment analysis on Chinese investor interactive platforms, utilizing relevant news, investor questions, and secretary responses as collective analytical data to achieve accurate analysis of investor sentiments. Specifically, to uncover latent information between different texts, we employ GPT for text comprehension and fusion. Addressing instances where questions involve multiple companies in the fusion text, we emloy prompt learning with RoBERTa to conduct sentiment analysis tailored to the questioning company. The accuracy of sentiment analysis reaches 97.61%, demonstrating an improvement of approximately 10% compared to the baseline.

In quantitative finance, the standard approach involves predicting stock returns to optimize asset allocation, aiming to maximize returns and minimize risks. This predict-then-optimize method traditionally focuses on minimizing prediction errors and optimizing a risk-adjusted objective function. Such a dichotomy often leads to suboptimal outcomes due to the misalignment between prediction and optimization goals [1]. In response to these challenges, while end-to-end deep portfolio architectures [2] – [4] have been proposed, their "black box" nature often fails to adapt to rapidly changing market conditions, resulting in inefficiencies and non-explainable outcomes [5].This paper introduces a novel network architecture that aligns the goals of minimizing prediction error with optimizing allocation by integrating a differentiable optimization layer [6]. We applied this architecture to Qlib, an AI-oriented quantitative investment platform, to simulate market environments and perform rigorous backtesting. Our results show significant enhancements in key financial metrics such as cumulative returns and average risk-adjusted objective values, surpassing traditional two-stage models.Additionally, we conducted portfolio experiments during the COVID-19 period to emphasize the importance of dynamic constraint adjustment in portfolio optimization and demonstrate our model’s effectiveness compared to end-to-end models.

Accurate identification of plant diseases is important for ensuring the safety of agricultural production. Convolutional neural networks (CNNs) and visual transformers (VTs) can extract effective representations of images and have been widely used for the intelligent recognition of plant disease images. However, CNNs have excellent local perception with poor global perception, and VTs have excellent global perception with poor local perception. This makes it difficult to further improve the performance of both CNNs and VTs on plant disease recognition tasks. In this paper, we propose a local and global feature-aware dual-branch network, named LGNet, for the identification of plant diseases. More specifically, we first design a dual-branch structure based on CNNs and VTs to extract the local and global features. Then, an adaptive feature fusion (AFF) module is designed to fuse the local and global features, thus driving the model to dynamically perceive the weights of different features. Finally, we design a hierarchical mixed-scale unit-guided feature fusion (HMUFF) module to mine the key information in the features at different levels and fuse the differentiated information among them, thereby enhancing the model's multiscale perception capability. Subsequently, extensive experiments were conducted on the AI Challenger 2018 dataset and the self-collected corn disease (SCD) dataset. The experimental results demonstrate that our proposed LGNet achieves state-of-the-art recognition performance on both the AI Challenger 2018 dataset and the SCD dataset, with accuracies of 88.74% and 99.08%, respectively.

Tomatoes are among the most extensively grown and consumed crops worldwide, but tomato production can be greatly reduced due to various diseases. Plant diseases show different symptoms at different stages. In addition, there are similarities in the symptoms of different types of plant diseases, which hinder the recognition of diseases by existing deep learning models. Traditional convolutional neural network (CNN) models for disease recognition have a large number of parameters that require high computational resources. To overcome these challenges, we propose a lightweight CNN model named LSGNet (lightweight sandglass network) for tomato disease identification. The LSGNet backbone consists of the sandglass with efficient channel attention (SGECA) and the position aware circular convolution sandglass (ParcSG) modules. The SGECA module reduces the interference of complex environments and thus focuses on extracting useful feature information. The ParcSG module has a global receptive field, which provides more detailed feature information on disease recognition. The results show that the recognition accuracies are 92.37%, 94.32%, 89.64%, 92.70%, 94.43%, 90.97%, 89.42%, 92.98%, 89.58%, and 95.54% for AlexNet, ResNet50, VGG16, MobileNetV3-Large, ShuffleNetV2-1×, EfficientNetV2-Small, ViT-Base, MobileViT-Small, Swin-Tiny, and LSGNet. Therefore, LSGNet has higher accuracy in recognizing tomato diseases compared to other classical models. In addition, LSGNet uses 0.75 million parameters. Compared to the lightweight CNN model MobileNetV3-Large, it only has 18% of the parameters. As a whole, the advantages of LSGNet in efficiency and lightweight structure would make it a useful tool for tomato disease recognition on mobile or embedded devices.

Pests are a major threat to the security of global agricultural production. Therefore, accurate identification of pests is vital for farmers to increase production and the associated income. In recent years, convolutional neural networks (CNNs) have become a mainstream method for pest identification. However, existing CNN-based approaches have a limitation due to the lack of key diverse feature representations, making it difficult to improve their recognition performance in large scale pest identification. To address the above limitation, we propose the hierarchical complementary network (HCNet) to capture pest feature representations and perform complementary fusion for obtaining hierarchical complementary information. Specifically, we first use a " shallow to deep " strategy to capture the hierarchical representations of the pest images. We then propose a spatial feature discrimination (SFD) module, which captures the key information in the hierarchical representations by boosting the spatial features of the current phase and suppressing the spatial features of the next phase. Finally, we design coordinate attention-guided feature complementary (CAFC) modules to fusion complementary information between features extracted from the SFD modules. Subsequently, we conduct experiments on the large scale pest dataset IP102. Without bells and whistles, the experimental results show that the proposed HCNet (ConvNext-B) achieves 75.36% accuracy on the test set, outperforming the existing state-of-the-art pest identification methods. Moreover, the proposed HCNet outperforms other state-of-the-art methods on different backbone networks. It will have a positive impact on the development of large scale pest identification methods.

Breast cancer is one of the most prevalent cancers in women. Reliable pathology identification can help histopathologists make accurate diagnosis of breast cancer but require specialized histopathological knowledge and a significant amount of manpower and medical resources. In this study, we fuse the coordinated attention mechanism to enhance the image texture analysis capability of the DenseNet, and build the CA-BreastNet model to classify microscopic histopathological images of specific types of breast cancers in the BreakHis dataset. More crucially, convolutional decision trees based on the specialized enhanced classifying strategy(SECS) are built to increase the overall accuracy of the network by reducing the model's accuracy restriction imposed by dataset structures. The related experimental results show that our network has strong performance and the SECS offers researchers reliable and effective performance enhancement guidelines. The accuracy of the convolutional decision trees reaches 99.75% for binary classification and 95.69% for eight-class classification, which means our model and strategy will be useful in the field of automatic diagnosis of breast cancer.

Market making (MM) is an important research topic in quantitative finance, the agent needs to continuously optimize ask and bid quotes to provide liquidity and make profits. The limit order book (LOB) contains information on all active limit orders, which is an essential basis for decisionmaking. The modeling of evolving, high-dimensional and low signal-to-noise ratio LOB data is a critical challenge. Traditional MM strategy relied on strong assumptions such as price process, order arrival process etc. Previous reinforcement learning (RL) works handcrafted market features, which is insufficient to represent the market. This paper proposes a RL agent for market making with LOB data. We leverage a neural network with convolutional filters and attention mechanism (Attn-LOB) for feature extraction from LOB. We design a new continuous action space and a hybrid reward function for the MM task. Finally, we conduct comprehensive experiments on latency and interpretability, showing that our agent has good applicability.

Introduction:Corn is one of the world's essential crops, and the presence of corn diseases significantly affects both the yield and quality of corn. Accurate identification of corn diseases in real time is crucial to increasing crop yield and improving farmers' income. However, in real-world environments, the complexity of the background, irregularity of the disease region, large intraclass variation, and small interclass variation make it difficult for most convolutional neural network models to achieve disease recognition under such conditions. Additionally, the low accuracy of existing lightweight models forces farmers to compromise between accuracy and real-time.Methods:To address these challenges, we propose FCA-EfficientNet. Building upon EfficientNet, the fully-convolution-based coordinate attention module allows the network to acquire spatial information through convolutional structures. This enhances the network's ability to focus on disease regions while mitigating interference from complex backgrounds. Furthermore, the adaptive fusion module is employed to fuse image information from different scales, reducing interference from the background in disease recognition. Finally, through multiple experiments, we have determined the network structure that achieves optimal performance.Results:Compared to other widely used deep learning models, this proposed model exhibits outstanding performance in terms of accuracy, precision, recall, and F1 score. Furthermore, the model has a parameter count of 3.44M and Flops of 339.74M, which is lower than most lightweight network models. We designed and implemented a corn disease recognition application and deployed the model on an Android device with an average recognition speed of 92.88ms, which meets the user's needs.Discussion:Overall, our model can accurately identify corn diseases in realistic environments, contributing to timely and effective disease prevention and control.

Since inflation of commodities is becoming more and more severe recently caused by many macro events, such as COVID-19 and Russian-Ukrainian conflict, systemic risk of commodity futures market is getting more attention from academic and industrial areas. Instead of using external factors to explain this risk as previous researches, we explain it by internal topology and structures of commodity futures market. This method helps us understand its key driving factors and their different impact to Chinese and international commodity futures markets.

Accurate identification of soybean leaf disease is of utmost importance for its cultivation and fine management, as it is a critical factor contributing to the decreased quality and yield of soybean. Nevertheless, several existing studies on the identification of soybean leaf disease encounter the challenge of striking a balance between model expressiveness and practical applicability. To address such issue, we propose a well-designed two-stage feature aggregation network framework (TFANet). The main component of it is the two-stage feature aggregation (TFA) module, which is designed to achieve great feature representation capability by aggregating feature information from different convolutional layers in two stages. Meanwhile, TFA module incorporates the efficient channel attention to extract crucial disease information. Afterwards, to limit the loss of feature information and expand the receptive field, a dilated convolution-based feature fusion (DCFF) module is developed. Additionally, the InceptionC module is introduced to further achieve better performance. The experimental results demonstrated that TFANet obtained an accuracy of 98.18% and an F1 scores of 98.39%, with merely 1.18 M parameters. In comparison with some classic convolutional neural network models, TFANet showed remarkable superiority in multiple evaluation metrics. Hence, TFANet has great potential to be used in practical soybean leaf dis-ease identification. (c) 2023 The Author(s). Published by Elsevier B.V. on behalf of King Saud University. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Introduction Tobacco brown spot disease caused by Alternaria fungal species is a major threat to tobacco growth and yield. Thus, accurate and rapid detection of tobacco brown spot disease is vital for disease prevention and chemical pesticide inputs. Methods Here, we propose an improved YOLOX-Tiny network, named YOLO-Tobacco, for the detection of tobacco brown spot disease under open-field scenarios. Aiming to excavate valuable disease features and enhance the integration of different levels of features, thereby improving the ability to detect dense disease spots at different scales, we introduced hierarchical mixed-scale units (HMUs) in the neck network for information interaction and feature refinement between channels. Furthermore, in order to enhance the detection of small disease spots and the robustness of the network, we also introduced convolutional block attention modules (CBAMs) into the neck network. Results As a result, the YOLO-Tobacco network achieved an average precision (AP) of 80.56% on the test set. The AP was 3.22%, 8.99%, and 12.03% higher than that obtained by the classic lightweight detection networks YOLOX-Tiny network, YOLOv5-S network, and YOLOv4-Tiny network, respectively. In addition, the YOLO-Tobacco network also had a fast detection speed of 69 frames per second (FPS). Discussion Therefore, the YOLO-Tobacco network satisfies both the advantages of high detection accuracy and fast detection speed. It will likely have a positive impact on early monitoring, disease control, and quality assessment in diseased tobacco plants.

针对多数卷积神经网络模型计算资源消耗多、 占用内存大等问题,提出了一种基于轻量化卷积神经网络的番茄病害图像识别方法FTL-MobileNet.该模型通过引入Focal Loss代替交叉熵损失函数(Cross Entropy Loss),聚焦于难分类的样本,接着在全连接层添加Dropout层,防止过拟合,将MobileNetV2在ImageNet训练好的权重参数迁移到改进模型中.选用常见的几种网络模型进行对比实验.结果表明,FTL-MobileNet相比于其他模型收敛更快、 泛化能力更好、 单张图片识别耗时更低、 识别精度更高,在测试集上的平均准确率达到了99.87％,且训练好的模型仅8.74 MB.

Topic modeling extracts useful potential topics that reflect market information from massive financial news and is widely used in data mining and economic research. Traditional topic modeling approaches such as Latent Dirichlet Allocation (LDA) and Non-Negative Matrix Factorization (NMF) lack semantic information, and short texts have feature sparse problems. We develop a topic clustering model based on BERT-LDA joint embedding that takes both contextual semantics and thematic narrative into account. We cluster document embeddings with the HDBSCAN algorithm and utilize a class-based TF-IDF (c-TF-IDF) method to create topic representations. Empirical results show that the BERT-LDA model is competitive compared with traditional and single topic models. It generates coherent topic words that are dissimilar to each other.

The deep learning model has achieved good results in index prediction, but most models only use structured data and cannot effectively predict mutation data. In order to better predict the sudden change data effectively, this paper proposes a new trend prediction model SEN_LSTM, which takes unstructured text data as a supplementary input and improves the performance of the model.