Winter wheat ear counting based on improved YOLOv7x and Kalman filter tracking algorithm with video streaming.

Accurate and real-time field wheat ear counting is of great significance for wheat yield prediction, genetic breeding and optimized planting management. In order to realize wheat ear detection and counting under the large-resolution Unmanned Aerial Vehicle (UAV) video, Space to depth (SPD) module was added to the deep learning model YOLOv7x. The Normalized Gaussian Wasserstein Distance (NWD) Loss function is designed to create a new detection model YOLOv7xSPD. The precision, recall, F1 score and AP of the model on the test set are 95.85%, 94.71%, 95.28%, and 94.99%, respectively. The AP value is 1.67% higher than that of YOLOv7x, and 10.41%, 39.32%, 2.96%, and 0.22% higher than that of Faster RCNN, SSD, YOLOv5s, and YOLOv7. YOLOv7xSPD is combined with the Kalman filter tracking and the Hungarian matching algorithm to establish a wheat ear counting model with the video flow, called YOLOv7xSPD Counter, which can realize real-time counting of wheat ears in the field. In the video with a resolution of 3840×2160, the detection frame rate of YOLOv7xSPD Counter is about 5.5FPS. The counting results are highly correlated with the ground truth number (R2 = 0.99), and can provide model basis for wheat yield prediction, genetic breeding and optimized planting management.

Effects of different ground segmentation methods on the accuracy of UAV-based canopy volume measurements.

The nonuniform distribution of fruit tree canopies in space poses a challenge for precision management. In recent years, with the development of Structure from Motion (SFM) technology, unmanned aerial vehicle (UAV) remote sensing has been widely used to measure canopy features in orchards to balance efficiency and accuracy. A pipeline of canopy volume measurement based on UAV remote sensing was developed, in which RGB and digital surface model (DSM) orthophotos were constructed from captured RGB images, and then the canopy was segmented using U-Net, OTSU, and RANSAC methods, and the volume was calculated. The accuracy of the segmentation and the canopy volume measurement were compared. The results show that the U-Net trained with RGB and DSM achieves the best accuracy in the segmentation task, with mean intersection of concatenation (MIoU) of 84.75% and mean pixel accuracy (MPA) of 92.58%. However, in the canopy volume estimation task, the U-Net trained with DSM only achieved the best accuracy with Root mean square error (RMSE) of 0.410 m3, relative root mean square error (rRMSE) of 6.40%, and mean absolute percentage error (MAPE) of 4.74%. The deep learning-based segmentation method achieved higher accuracy in both the segmentation task and the canopy volume measurement task. For canopy volumes up to 7.50 m3, OTSU and RANSAC achieve an RMSE of 0.521 m3 and 0.580 m3, respectively. Therefore, in the case of manually labeled datasets, the use of U-Net to segment the canopy region can achieve higher accuracy of canopy volume measurement. If it is difficult to cover the cost of data labeling, ground segmentation using partitioned OTSU can yield more accurate canopy volumes than RANSAC.

Large field-of-view pine wilt disease tree detection based on improved YOLO v4 model with UAV images.

Introduction: Pine wilt disease spreads rapidly, leading to the death of a large number of pine trees. Exploring the corresponding prevention and control measures for different stages of pine wilt disease is of great significance for its prevention and control. Methods: To address the issue of rapid detection of pine wilt in a large field of view, we used a drone to collect multiple sets of diseased tree samples at different times of the year, which made the model trained by deep learning more generalizable. This research improved the YOLO v4(You Only Look Once version 4) network for detecting pine wilt disease, and the channel attention mechanism module was used to improve the learning ability of the neural network. Results: The ablation experiment found that adding the attention mechanism SENet module combined with the self-designed feature enhancement module based on the feature pyramid had the best improvement effect, and the mAP of the improved model was 79.91%. Discussion: Comparing the improved YOLO v4 model with SSD, Faster RCNN, YOLO v3, and YOLO v5, it was found that the mAP of the improved YOLO v4 model was significantly higher than the other four models, which provided an efficient solution for intelligent diagnosis of pine wood nematode disease. The improved YOLO v4 model enables precise location and identification of pine wilt trees under changing light conditions. Deployment of the model on a UAV enables large-scale detection of pine wilt disease and helps to solve the challenges of rapid detection and prevention of pine wilt disease.

YOLO SSPD: a small target cotton boll detection model during the boll-spitting period based on space-to-depth convolution.

Introduction: Cotton yield estimation is crucial in the agricultural process, where the accuracy of boll detection during the flocculation period significantly influences yield estimations in cotton fields. Unmanned Aerial Vehicles (UAVs) are frequently employed for plant detection and counting due to their cost-effectiveness and adaptability. Methods: Addressing the challenges of small target cotton bolls and low resolution of UAVs, this paper introduces a method based on the YOLO v8 framework for transfer learning, named YOLO small-scale pyramid depth-aware detection (SSPD). The method combines space-to-depth and non-strided convolution (SPD-Conv) and a small target detector head, and also integrates a simple, parameter-free attentional mechanism (SimAM) that significantly improves target boll detection accuracy. Results: The YOLO SSPD achieved a boll detection accuracy of 0.874 on UAV-scale imagery. It also recorded a coefficient of determination (R2) of 0.86, with a root mean square error (RMSE) of 12.38 and a relative root mean square error (RRMSE) of 11.19% for boll counts. Discussion: The findings indicate that YOLO SSPD can significantly improve the accuracy of cotton boll detection on UAV imagery, thereby supporting the cotton production process. This method offers a robust solution for high-precision cotton monitoring, enhancing the reliability of cotton yield estimates.

Prediction of wheat SPAD using integrated multispectral and support vector machines.

Rapidly obtaining the chlorophyll content of crop leaves is of great significance for timely diagnosis of crop health and effective field management. Multispectral imagery obtained from unmanned aerial vehicles (UAV) is being used to remotely sense the SPAD (Soil and Plant Analyzer Development) values of wheat crops. However, existing research has not yet fully considered the impact of different growth stages and crop populations on the accuracy of SPAD estimation. In this study, 300 materials from winter wheat natural populations in Xinjiang, collected between 2020 to 2022, were analyzed. UAV multispectral images were obtained in the experimental area, and vegetation indices were extracted to analyze the correlation between the selected vegetation indices and SPAD values. The input variables for the model were screened, and a support vector machine (SVM) model was constructed to estimate SPAD values during the heading, flowering, and filling stages under different water stresses. The aim was to provide a method for the rapid acquisition of winter wheat SPAD values. The results showed that the SPAD values under normal irrigation were higher than those under water restriction. Multiple vegetation indices were significantly correlated with SPAD values. In the prediction model construction of SPAD, the different models had high estimation accuracy under both normal irrigation and water limitation treatments, with correlation coefficients of predicted and measured values under normal irrigation in different environments the value of r from 0.59 to 0.81 and RMSE from 2.15 to 11.64, compared to RE from 0.10% to 1.00%; and under drought stress in different environments, correlation coefficients of predicted and measured values of r was 0.69-0.79, RMSE was 2.30-12.94, and RE was 0.10%-1.30%. This study demonstrated that the optimal combination of feature selection methods and machine learning algorithms can lead to a more accurate estimation of winter wheat SPAD values. In summary, the SVM model based on UAV multispectral images can rapidly and accurately estimate winter wheat SPAD value.

The specifics of the Galois field GF(257) and its use for digital signal processing.

An algorithm of digital logarithm calculation for the Galois field G F ( 257 ) is proposed. It is shown that this field is coupled with one of the most important existing standards that uses a digital representation of the signal through 256 levels. It is shown that for this case it is advisable to use the specifics of quasi-Mersenne prime numbers, representable in the form p = 2 n + 1 , which includes the number 257. For fields G F ( 2 n + 1 ) , an alternating encoding can be used, in which non-zero elements of the field are displayed through binary characters corresponding to the numbers + 1 and - 1. In such an encoding, multiplying a field element by 2 is reduced to a quasi-cyclic permutation of binary symbols (the permuted symbol changes sign). Proposed approach makes it possible to significantly simplify the design of computing devices for calculation of digital logarithm and multiplication of numbers modulo 257. A concrete scheme of a device for digital logarithm calculation in this field is presented. It is also shown that this circuit can be equipped with a universal adder modulo an arbitrary number, which makes it possible to implement any operations in the field under consideration. It is shown that proposed digital algorithm can also be used to reduce 256-valued logic operations to algebraic form. It is shown that the proposed approach is of significant interest for the development of UAV on-board computers operating as part of a group.

Risk assessment of environmental and bystander exposure from agricultural unmanned aerial vehicle sprayers in golden coconut plantations: Effects of droplet size and spray volume.

Unmanned aerial vehicle (UAV) sprayers are widely utilized in commercial aerial application of plant protection products (PPPs) in East Asian countries due to their high flexibility, high efficiency and low cost, but spray drift can lead to low utilization of UAV sprayers application, environmental pollution and bystander exposure risk. Droplet size and spray volume are critical factors affecting spray drift. Currently, the high temperature and humidity environment under the influence of the tropical monsoon climate brings new challenges for UAV sprayers. Therefore, in this study, pesticides were simulated with seduction red solution, and spraying trials were conducted using the DJI commercial T40 UAV sprayers for a typical tropical crop, coconut. In this study, the spray drift distribution of droplets on the ground and in the air, as well as the bystander exposure risk, were comparatively analyzed using droplet size (VF, M, and C) and spray volume (75 L/hm2 and 60 L/hm2) as trial variables. The results indicated that the spray drift characteristics of UAV sprayers were significantly affected by droplet size and spray volume. The spray drift percentage was negatively correlated with the downwind distance and the sampling height. The smaller the droplet size, the farther the buffer zone distance, up to more than 30 m, and the cumulative drift percentage is also greater, resulting in a significant risk of spray drift. The reduction in spray volume helped to reduce the spray drift, and the cumulative drift percentage was reduced by 73.87 % with a droplet size of M. The region of the body where spray drift is deposited the most on bystanders is near chest height. This study provides a reference for the rational and safe use of multirotor UAV sprayers application operations in the tropics and enriches the spray drift database in the tropics.

Determination of the effective swath of a plant protection UAV adapted to mist nozzles in mountain Nangguo pear orchards.

Plant protection unmanned aerial vehicles (UAVs) have become popular in mountain orchards, but due to the differences in planting structures, the chances of heavy spraying, missed spraying and pesticide drift are increasing. To mitigate the adverse effects of these phenomena, it is necessary to clarify the effective deposition range of aerial spray droplets. This study proposed an effective spray swath determination method for the effective spraying range of mountainous orchards with UAVs equipped with a mist nozzle (bilateral 1% coverage). This approach focused on exploring the effects of flight height (unidirectional flight modes of 2, 3 and 4 m), spray nozzle atomization performance (reciprocating flight modes of 20, 30 and 40 µm) and flight route (treetop flying and inter-row flying) on the spraying range in a mountain setting. In addition, the study analysed the relationship between the droplet-size spectrum and the effective swath position. The results showed that it is feasible to use the bilateral 1% coverage evaluation method to determine the effective spray swath of a UAV adapted with a mist nozzle for aerial operation in a mountainous Nangguo Pear orchard. With the increase in UAV flight height (2-4 m), the effective unidirectional spray swath also increased, and with the increase in atomization level (20-40 µm), the effective reciprocating spray swath showed a decreasing trend. Moreover, the average effective swath width measured by the UAV for treetop flight was greater than that measured for inter-row flight. The study also found that the proportion of small droplets (droplet size less than 100 µm) below the UAV route was lower (approximately 50%) than along the sides of the route (approximately 80%), and the spray swath was not symmetrically distributed along the flight route but shifted laterally by approximately 3 to 4 m in the downhill direction.

European beech spring phenological phase prediction with UAV-derived multispectral indices and machine learning regression.

Acquiring phenological event data is crucial for studying the impacts of climate change on forest dynamics and assessing the risks associated with the early onset of young leaves. Large-scale mapping of forest phenological timing using Earth observation (EO) data could enhance our understanding of these processes through an added spatial component. However, translating traditional ground-based phenological observations into reliable ground truthing for training and validating EO mapping applications remains challenging. This study explored the feasibility of predicting high-resolution phenological phase data for European beech (Fagus sylvatica) using unoccupied aerial vehicle (UAV)-based multispectral indices and machine learning. Employing a comprehensive feature selection process, we identified the most effective sensors, vegetation indices, training data partitions, and machine learning models for phenological phase prediction. The model that performed best and generalized well across various sites utilized Green Chromatic Coordinate (GCC) and Generalized Additive Model (GAM) boosting. The GCC training data, derived from the radiometrically calibrated visual bands of a multispectral sensor, were predicted using uncalibrated RGB sensor data. The final GCC/GAM boosting model demonstrated capability in predicting phenological phases on unseen datasets within a root mean squared error threshold of 0.5. This research highlights the potential interoperability among common UAV-mounted sensors, particularly the utility of readily available, low-cost RGB sensors. However, considerable limitations were observed with indices that implement the near-infrared band due to oversaturation. Future work will focus on adapting models to better align with the ICP Forests phenological flushing stages.

HeMoDU: High-Efficiency Multi-Object Detection Algorithm for Unmanned Aerial Vehicles on Urban Roads.

Unmanned aerial vehicle (UAV)-based object detection methods are widely used in traffic detection due to their high flexibility and extensive coverage. In recent years, with the increasing complexity of the urban road environment, UAV object detection algorithms based on deep learning have gradually become a research hotspot. However, how to further improve algorithmic efficiency in response to the numerous and rapidly changing road elements, and thus achieve high-speed and accurate road object detection, remains a challenging issue. Given this context, this paper proposes the high-efficiency multi-object detection algorithm for UAVs (HeMoDU). HeMoDU reconstructs a state-of-the-art, deep-learning-based object detection model and optimizes several aspects to improve computational efficiency and detection accuracy. To validate the performance of HeMoDU in urban road environments, this paper uses the public urban road datasets VisDrone2019 and UA-DETRAC for evaluation. The experimental results show that the HeMoDU model effectively improves the speed and accuracy of UAV object detection.

Downlink Transmissions of UAV-RIS-Assisted Cell-Free Massive MIMO Systems: Location and Trajectory Optimization.

In this paper, we investigate a cell-free massive multiple-input multiple-output (CF-mMIMO) system with a reconfigurable intelligent surface (RIS) carried by an unmanned aerial vehicle (UAV), called the UAV-RIS. Compared with the RIS located on the ground, the UAV-RIS has a wider coverage that can reflect all signals from access points (APs) and user equipment (UE). By correlating the UAV location with the Rician K-factor, we derive a closed-form approximation of the UE achievable downlink rate. Based on this, we obtain the optimal UAV location and RIS phase shift that can maximize the UE sum rate through an alternating optimization method. Simulation results have verified the accuracy of the derived approximation and shown that the UE sum rate can be significantly improved with the obtained optimal UAV location and RIS phase shift. Moreover, we find that with a uniform UE distribution, the UAV-RIS should fly to the center of the system, while with an uneven UE distribution, the UAV-RIS should fly above the area where UEs are gathered. In addition, we also design the best trajectory for the UAV-RIS to fly from its initial location to the optimal destination while maintaining the maximum UE sum rate per time slot during the flight.

Unmanned Aerial Vehicle-Assisted Terahertz-Visible Light Communication Systems: An In-Depth Performance Analysis.

This paper investigates the performance of dual-hop unmanned aerial vehicle (UAV)-assisted communication channels, employing a decode-and-forward (DF) relay architecture. The system leverages terahertz (THz) communication for the primary hop and visible light communication (VLC) for the secondary hop. We conduct an in-depth analysis by deriving closed-form expressions for the end-to-end (E2E) bit error rate (BER). Additionally, we use a Monte Carlo simulation approach to generate best-fitting curves, validating our analytical expressions. A performance evaluation through BER and outage probability metrics demonstrates the effectiveness of the proposed system. Specifically, our results indicate that the proposed system outperforms Free-Space Optics (FSO)-VLC and Radio-Frequency (RF)-VLC at a higher signal-to-noise ratio (SNR). The results of this study provide valuable insights into the feasibility and limitations of UAV-assisted THz-VLC communication systems.

Surface Defect-Extended BIM Generation Leveraging UAV Images and Deep Learning.

Defect inspection of existing buildings is receiving increasing attention for digitalization transfer in the construction industry. The development of drone technology and artificial intelligence has provided powerful tools for defect inspection of buildings. However, integrating defect inspection information detected from UAV images into semantically rich building information modeling (BIM) is still challenging work due to the low defect detection accuracy and the coordinate difference between UAV images and BIM models. In this paper, a deep learning-based method coupled with transfer learning is used to detect defects accurately; and a texture mapping-based defect parameter extraction method is proposed to achieve the mapping from the image U-V coordinate system to the BIM project coordinate system. The defects are projected onto the surface of the BIM model to enrich a surface defect-extended BIM (SDE-BIM). The proposed method was validated in a defect information modeling experiment involving the No. 36 teaching building of Nantong University. The results demonstrate that the methods are widely applicable to various building inspection tasks.

Drone-Borne Magnetic Gradiometry in Archaeological Applications.

The use of magnetometers arranged in a gradiometer configuration offers a practical and widely used solution, particularly in archaeological applications where the sources of interest are generally shallow. Since magnetic anomalies due to archaeological remains often have low amplitudes, highly sensitive magnetic sensors are kept very close to the ground to reveal buried structures. However, the deployment of Unmanned Aerial Vehicles (UAVs) is increasingly becoming a reliable and valuable tool for the acquisition of magnetic data, providing uniform coverage of large areas and access to even very steep terrain, saving time and reducing risks. However, the application of a vertical gradiometer for drone-borne measurements is still challenging due to the instability of the system drone magnetometer in flight and noise issues due to the magnetic interference of the mobile platform or related to the oscillation of the suspended sensors. We present the implementation of a magnetic vertical gradiometer UAV system and its use in an archaeological area of Southern Italy. To reduce the magnetic and electromagnetic noise caused by the aircraft, the magnetometer was suspended 3m below the drone using ropes. A Continuous Wavelet Transform analysis of data collected in controlled tests confirmed that several characteristic power spectrum peaks occur at frequencies compatible with the magnetometer oscillations. This noise was then eliminated with a properly designed low-pass filter. The resulting drone-borne vertical gradient data compare very well with ground-based magnetic measurements collected in the same area and taken as a control dataset.

Dynamic UAV Deployment Scheme Based on Edge Computing for Forest Fire Scenarios.

This study investigates the dynamic deployment of unmanned aerial vehicles (UAVs) using edge computing in a forest fire scenario. We consider the dynamically changing characteristics of forest fires and the corresponding varying resource requirements. Based on this, this paper models a two-timescale UAV dynamic deployment scheme by considering the dynamic changes in the number and position of UAVs. In the slow timescale, we use a gate recurrent unit (GRU) to predict the number of future users and determine the number of UAVs based on the resource requirements. UAVs with low energy are replaced accordingly. In the fast timescale, a deep-reinforcement-learning-based UAV position deployment algorithm is designed to enable the low-latency processing of computational tasks by adjusting the UAV positions in real time to meet the ground devices' computational demands. The simulation results demonstrate that the proposed scheme achieves better prediction accuracy. The number and position of UAVs can be adapted to resource demand changes and reduce task execution delays.

An Improved Instance Segmentation Method for Fast Assessment of Damaged Buildings Based on Post-Earthquake UAV Images.

Quickly and accurately assessing the damage level of buildings is a challenging task for post-disaster emergency response. Most of the existing research mainly adopts semantic segmentation and object detection methods, which have yielded good results. However, for high-resolution Unmanned Aerial Vehicle (UAV) imagery, these methods may result in the problem of various damage categories within a building and fail to accurately extract building edges, thus hindering post-disaster rescue and fine-grained assessment. To address this issue, we proposed an improved instance segmentation model that enhances classification accuracy by incorporating a Mixed Local Channel Attention (MLCA) mechanism in the backbone and improving small object segmentation accuracy by refining the Neck part. The method was tested on the Yangbi earthquake UVA images. The experimental results indicated that the modified model outperformed the original model by 1.07% and 1.11% in the two mean Average Precision (mAP) evaluation metrics, mAPbbox50 and mAPseg50, respectively. Importantly, the classification accuracy of the intact category was improved by 2.73% and 2.73%, respectively, while the collapse category saw an improvement of 2.58% and 2.14%. In addition, the proposed method was also compared with state-of-the-art instance segmentation models, e.g., Mask-R-CNN and YOLO V9-Seg. The results demonstrated that the proposed model exhibits advantages in both accuracy and efficiency. Specifically, the efficiency of the proposed model is three times faster than other models with similar accuracy. The proposed method can provide a valuable solution for fine-grained building damage evaluation.

Optimizing UAV spray parameters to improve precise control of tobacco pests at different growth stages.

BACKGROUND: Unmanned aerial vehicles (UAVs) for pesticide application show promising potential in tobacco pest management. However, the impact of flight parameters on spray efficacy requires further investigation. Three field experiments were conducted from the rosette to the maturation stage of tobacco to systematically assess spray efficacy under varying flight heights, speeds, and application volumes. Using a multi-index weight analysis method, optimal operational parameter combinations for different tobacco growth stages were evaluated and compared with backpack electric sprayers. RESULTS: For the rosette stage, the recommended parameter is a flight speed of 5 m s-1, a flight height of 2 m, and a liquid application volume of 30 L hm-2; during the vigorous growth stage, the suggested parameter includes a flight speed of 3 m s-1, a flight height of 2 m, and a liquid application volume of 22.5 L hm-2. In the maturing stage, optimal parameter consists of a flight speed of 3 m s-1, a flight height of 3.5 m, and a liquid application volume of 30 L hm-2. Furthermore, UAV spraying achieves higher droplet deposition on both sides of tobacco leaves compared to traditional electric backpack sprayers. CONCLUSIONS: Adjusting UAV spraying parameters for different tobacco growth stages is crucial. These results can provide the methods for the precise control technology of tobacco pests at different growth stages. © 2024 Society of Chemical Industry.

Dynamics modeling and nonlinear attitude controller design for a rocket-type unmanned aerial vehicle.

This paper presents an altitude and attitude control system for a newly designed rocket-type unmanned aerial vehicle (UAV) propelled by a gimbal-based coaxial rotor system (GCRS) enabling thrust vector control (TVC). The GCRS is the only means of actuation available to control the UAV's orientation, and the flight dynamics identify the primary control difficulty as the highly nonlinear and tightly coupled control distribution problem. To address this, the study presents detailed derivations of attitude flight dynamics and a control strategy to track the desired attitude trajectory. First, a Proportional-Integral-Derivative (PID) control algorithm is developed based on the formulation of linear matrix inequality (LMI) to ensure robust stability and performance. Second, an optimization algorithm using the Levenberg-Marquardt (LM) method is introduced to solve the nonlinear inverse mapping problem between the control law and the actual actuator outputs, addressing the nonlinear coupled control input distribution problem of the GCRS. In summary, the main contribution is the proposal of a new TVC UAV system based on GCRS. The PID control algorithm and LM algorithm were designed to solve the distribution problem of the actuation model and confirm altitude and attitude tracking missions. Finally, to validate the flight properties of the rocket-type UAV and the performance of the proposed control algorithm, several numerical simulations were conducted. The results indicate that the tightly coupled control input nonlinear inverse problem was successfully solved, and the proposed control algorithm achieved effective attitude stabilization even in the presence of disturbances.

Research on detection method of photovoltaic cell surface dirt based on image processing technology.

In view of the reduced power generation efficiency caused by ash or dirt on the surface of photovoltaic panels, and the problems of heavy workload and low efficiency faced by manual detection, this study proposes a method to detect dust or dust on the surface of photovoltaic cells with the help of image processing technology to timely eliminate hidden dangers and improve power generation efficiency.This paper introduces image processing methods based on mathematical morphology, such as image enhancement, image sharpening, image filtering and image closing operation, which makes the image better highlight the target to be recognized. At the same time, it also solves the problem of uneven image binarization caused by uneven illumination in the process of image acquisition. By using the image histogram equalization, the gray level concentration area of the original image is opened or the gray level is evenly distributed, so that the dynamic range of the pixel gray level is increased, so that the image contrast or contrast is increased, the image details are clear, to achieve the purpose of enhancement. When identifying the target area, the method of calculating the proportion of the dirt area to the whole image area is adopted, and the ratio exceeding a certain threshold is judged as a fault. In addition, the improved A* path planning algorithm is adopted in this study, which greatly improves the efficiency of the unmanned aerial vehicle detection of photovoltaic cell dirt, saves time and resources, reduces operation and maintenance costs, and improves the operation and maintenance level of photovoltaic units.

Identifying rice field weeds from unmanned aerial vehicle remote sensing imagery using deep learning.

BACKGROUND: Rice field weed object detection can provide key information on weed species and locations for precise spraying, which is of great significance in actual agricultural production. However, facing the complex and changing real farm environments, traditional object detection methods still have difficulties in identifying small-sized, occluded and densely distributed weed instances. To address these problems, this paper proposes a multi-scale feature enhanced DETR network, named RMS-DETR. By adding multi-scale feature extraction branches on top of DETR, this model fully utilizes the information from different semantic feature layers to improve recognition capability for rice field weeds in real-world scenarios. METHODS: Introducing multi-scale feature layers on the basis of the DETR model, we conduct a differentiated design for different semantic feature layers. The high-level semantic feature layer adopts Transformer structure to extract contextual information between barnyard grass and rice plants. The low-level semantic feature layer uses CNN structure to extract local detail features of barnyard grass. Introducing multi-scale feature layers inevitably leads to increased model computation, thus lowering model inference speed. Therefore, we employ a new type of Pconv (Partial convolution) to replace traditional standard convolutions in the model. RESULTS: Compared to the original DETR model, our proposed RMS-DETR model achieved an average recognition accuracy improvement of 3.6% and 4.4% on our constructed rice field weeds dataset and the DOTA public dataset, respectively. The average recognition accuracies reached 0.792 and 0.851, respectively. The RMS-DETR model size is 40.8 M with inference time of 0.0081 s. Compared with three classical DETR models (Deformable DETR, Anchor DETR and DAB-DETR), the RMS-DETR model respectively improved average precision by 2.1%, 4.9% and 2.4%. DISCUSSION: This model is capable of accurately identifying rice field weeds in complex real-world scenarios, thus providing key technical support for precision spraying and management of variable-rate spraying systems.