Explore the vast range of titles available.

Chemical control using pesticides remains an essential component of crop pest and disease management, while precision pesticide application is a core element for achieving sustainable agriculture. Precision spraying technology—integrating UAV-based detection, real-time pesticide mixing, and adaptive variable-rate spraying—provides a critical pathway for sustainable crop protection by establishing a “perception-decision-execution” closed-loop framework.While previous reviews have predominantly focused on compartmentalized analyses of individual technologies (e.g., sensing or actuation), this study establishes a unified Perception-Decision-Execution (PDE) framework to, for the first time, quantitatively assess the synergistic interactions and systemic Bottlenecks across all three layers.This paper systematically reviews 168 core publications from 2013 to 2024, proposing for the first time and quantitatively assessing the synergistic effects of technologies within this closed-loop framework. The findings reveal that: (1) UAV-deep learning systems achieve pest identification accuracy rates of 89–94%, but this significantly declines to 60–70% under strong light or occlusion conditions; (2) Real-time mixing systems attain a mixing homogeneity coefficient (γ) > 85% for liquid pesticides, while for suspension concentrates (SCs), γ decreases to 70–75% due to particle sedimentation effects; (3) PWM-based variable-rate spraying reduces pesticide usage by 30–50% and off-target drift by > 30%, though sensor errors can cause positioning deviations of 0.3–0.8 m. Despite considerable promise, this integrated technology faces challenges in large-scale applications, including perception degradation under environmental disturbances, limitations in algorithm generalization, poor pesticide formulation adaptability in mixing, and system coordination issues. To overcome these barriers, this review proposes interdisciplinary solutions: (i) Deploying lightweight edge devices and pruned neural networks to address decision-making delays and enhance real-time responsiveness; (ii) Optimizing mixer structures (e.g., helical baffle angles) based on computational fluid dynamics (CFD) simulations to reduce dead zones and improve mixing homogeneity for SCs; (iii) Integrating multi-sensor technology for drift compensation to enhance UAV spraying stability. By integrating and optimizing these key technologies, the closed-loop framework holds significant potential to markedly improve pesticide utilization efficiency, minimize environmental impact, and offer a practical framework for achieving on-demand application, thereby advancing sustainable smart agriculture.

Precision PWM variable spray technology and target detection, identification, and localization technology are key to solving the pesticide waste associated with traditional constant application methods and to improving pesticide utilization for achieving precise application. To address the problems of high pesticide dosage, low application efficiency, and poor kale pest and disease control in traditional upland gap sprayers, a variable spray control system was designed in the study. The system utilizes binocular vision sensors to detect kale targets in the field in real time and achieves accurate pesticide application through pulse-width modulation technology. An improved target detection model based on YOLOv8n is presented, with experimental results showing a detection accuracy of up to 88.7% for field-grown kale. The system was also tested for accuracy-responsive variable spraying in recognition detection tasks, with a 0.2% reduction in the central atomized deposition density coefficient of variation (CV) compared to constant spraying. A flow on/off test model was designed for the solenoid valve duty cycle, determining the correlation decision coefficient for spraying. The correlation coefficient of the flow model exceeded 0.9958 when the duty cycle was in the range of 20–90%, and the actual and theoretical flow rates at the spray terminals were strongly linearly correlated, with a maximum error of only 4.1%. The spraying effect of the system was evaluated through field tests. The results show that the theoretical spray volume of the variable spray control system aligns well with the actual spray volume. In field atomization deposition tests, compared with constant-rate spraying, the target center atomization density in variable spraying mode reached 34.42%. Although droplet deposition and coverage around the crop were slightly reduced, pest and disease control around kale remained effective. In addition, the variable-rate spraying control system further improved pesticide utilization, with a maximum pesticide savings of 26.58%. This study demonstrates the feasibility of binocular vision sensor-guided spraying operations in field environments and provides a reference for its application in field pest control.

QR codes are extensively utilized in multiple domains including product management, security authentication, and intelligent Internet of Things. However, current QR code detection models can be hindered by complex backgrounds, low real-time performance, and high resource consumption. To solve this problem, based on the You Only Look Once version 5 (YOLOv5) algorithm, we propose a lightweight real-time QR code detection algorithm called EBP-YOLOv5. This work introduces an Efficient Channel Attention mechanism in the backbone network to strengthen the network feature extraction capability. Bi-directional Feature Pyramid Network is utilized to replace the PANet module to enhance the network’s ability to capture features along the path. Additionally, EBP-YOLOv5 undergoes lightweight operation by reducing the number of model parameters through sparse training. L1 regularization is incorporated into the loss function to prune the weights of the Batch Normalization layer. Finally, distillation learning is applied to enhance the accuracy of the network model. Extensive evaluations on a natural scene QR code dataset demonstrate that EBP-YOLOv5 achieves superior performance. Compared to the original YOLOv5s algorithm, EBP-YOLOv5 enhances the average accuracy by 4.6% while reducing the model size to only 9MB. This significantly reduces the parameter count and computational load while maintaining a high average accuracy of 97%. Meanwhile, EBP-YOLOv5 outperforms other lightweight models in terms of parameter count, computational complexity, detection accuracy, and model size, and is more suitable for deployment on edge devices.

To solve the problems of difficult quantification of complex driving scenes and unclear classification, a method of complex measurement and scene classification was proposed. Based on the Bayesian network, the posterior probability distribution was obtained, the variable weights were determined by information entropy theory and BP neural network, and the gravitational model was improved so that the complex metric model of the driving scene was established, the static and dynamic complexity of the scene was quantified respectively, and a weighted fusion of the two was conducted. The K-means clustering method was used to divide the driving scenario into three categories, i.e., simple scenario, medium complex scenario, and complex scenario, and the rationality of the method was verified by experiments. This scenario complex metric method can provide a reference for studying the complex metrics and scene classification of smart vehicle test scenarios.

This paper proposes an improved YOLOv5 model based on the EfficientNet‐B0 backbone, EIoU loss function, and dynamic convolution operator, which was evaluated using the NEU‐DET dataset. The results show that compared with the traditional YOLOv5 model, the proposed improvement achieves a higher detection accuracy with fewer model parameters, resulting in a more lightweight architecture. Using the EfficientNet‐B0 backbone not only enhances the detection performance but also reduces the parameter count of the model, thereby, effectively improving computational efficiency. The enhanced EIoU loss function provided a more accurate assessment of the quality of the detected bounding boxes than the traditional CIoU loss function. Additionally, the dynamic convolution operator utilized in the neck module enhances the representation capability of the feature map while reducing computational overhead. The simulation results demonstrate that the proposed improved model achieves higher mean average precision (mAP) values and requires fewer parameters for the NEU‐DET dataset, thereby, highlighting its promising practical applications.

Background Long noncoding RNAs (lncRNAs) participate in the carcinogenesis of many different cancers. This study aimed to detect expression of lncRNA CTA‐941F9.9 in colorectal cancer tissues compared with matched nontumorous adjacent tissues (NATs). Moreover, we investigated whether this molecule is able to influence carcinogenesis in colorectal cancer (CRC). Methods Colorectal cancer tissues and NATs from two cohorts of patients were examined. Quantitative PCR was performed to quantify levels of CTA‐941F9.9 expression in these samples. The association between CTA‐941F9.9 expression and clinicopathological features, including receiver operating characteristic (ROC) curves, was also analyzed to evaluate the diagnostic value of CTA‐941F9.9 in CRC. Potential effects of lncRNA CTA‐941F9.9 on CRC cells were assessed via autophagy, transwell assay, CCK8 assays, and flow cytometry. Results Our experimental results showed lncRNA CTA‐941F9.9 to be significantly downregulated in CRC tissues in both cohorts, with areas under the ROC curve (AUC) of 0.802 and 0.876. However, no significant correlations between CTA‐941F9.9 expression levels and clinicopathological characteristics or patient outcomes were observed. We also found that CTA‐941F9.9 promotes autophagy in CRC cell lines but no significant function of CTA‐941F9.9 in regulating cancer cell proliferation or migration. Conclusions LncRNA CTA‐941F9.9 is frequently downregulated in CRC compared with NATs and might play an important role in CRC carcinogenesis.

For reasons such as privacy, there are use cases for language models at the edge. This has given rise to small language models targeted for deployment in resource-constrained devices where energy efficiency is critical. Spiking neural networks (SNNs) offer a promising solution due to their energy efficiency, and there are already works on realizing transformer-based models on SNNs. However, key operations like softmax and layer normalization (LN) are difficult to implement on neuromorphic hardware, and many of these early works sidestepped them. To address these challenges, we introduce Sorbet, a transformer-based spiking language model that is more neuromorphic hardware-compatible. Sorbet incorporates a novel shifting-based softmax called PTsoftmax and a Bit Shifting PowerNorm (BSPN), both designed to replace the respective energy-intensive operations. By leveraging knowledge distillation and model quantization, Sorbet achieved a highly compressed binary weight model that maintains competitive performance while achieving \\(27.16\\times\\) energy savings compared to BERT. We validate Sorbet through extensive testing on the GLUE benchmark and a series of ablation studies, demonstrating its potential as an energy-efficient solution for language model inference. Our code is publicly available at \\href{https://github.com/Kaiwen-Tang/Sorbet}{https://github.com/Kaiwen-Tang/Sorbet}

For reasons such as privacy, there are use cases for language models at the edge. This has given rise to small language models (SLMs) targeted for deployment in resource-constrained devices where energy efficiency is a significant concern. Spiking neural networks (SNNs) offer a promising solution due to their energy efficiency, and there are already works on realizing transformer-based models on SNNs. However, key operations like softmax and layer normalization (LN) are difficult to implement on neuromorphic hardware, and many of these early works sidestepped them. To address these challenges, we introduce Sorbet, a transformer-based spiking language model that is more neuromorphic hardware-compatible. Sorbet incorporates a novel shifting-based softmax called PTsoftmax and a power normalization method using bit-shifting (BSPN), both designed to replace the respective energy-intensive operations. By leveraging knowledge distillation and model quantization, Sorbet achieved a highly compressed binary weight model that maintains competitive performance while significantly reducing energy consumption. We validate Sorbet's effectiveness through extensive testing on the GLUE benchmark and a series of ablation studies, demonstrating its potential as an energy-efficient solution for language model inference.

Hyperdimensional computing (HDC) is a method to perform classification that uses binary vectors with high dimensions and the majority rule. This approach has the potential to be energy-efficient and hence deemed suitable for resource-limited platforms due to its simplicity and massive parallelism. However, in order to achieve high accuracy, HDC sometimes uses hypervectors with tens of thousands of dimensions. This potentially negates its efficiency advantage. In this paper, we examine the necessity of such high dimensions and conduct a detailed theoretical analysis of the relationship between hypervector dimensions and accuracy. Our results demonstrate that as the dimension of the hypervectors increases, the worst-case/average-case HDC prediction accuracy with the majority rule decreases. Building on this insight, we develop HDC models that use binary hypervectors with dimensions orders of magnitude lower than those of state-of-the-art HDC models while maintaining equivalent or even improved accuracy and efficiency. For instance, on the MNIST dataset, we achieve 91.12% HDC accuracy in image classification with a dimension of only 64. Our methods perform operations that are only 0.35% of other HDC models with dimensions of 10,000. Furthermore, we evaluate our methods on ISOLET, UCI-HAR, and Fashion-MNIST datasets and investigate the limits of HDC computing.

Vertical-domain large language models (LLMs) play a crucial role in specialized scenarios such as finance, healthcare, and law; however, their training often relies on large-scale annotated data and substantial computational resources, impeding rapid development and continuous iteration. To address these challenges, we introduce the Collaborative Editable Model (CoEM), which constructs a candidate knowledge pool from user-contributed domain snippets, leverages interactive user-model dialogues combined with user ratings and attribution analysis to pinpoint high-value knowledge fragments, and injects these fragments via in-context prompts for lightweight domain adaptation. With high-value knowledge, the LLM can generate more accurate and domain-specific content. In a financial information scenario, we collect 15k feedback from about 120 users and validate CoEM with user ratings to assess the quality of generated insights, demonstrating significant improvements in domain-specific generation while avoiding the time and compute overhead of traditional fine-tuning workflows.