Real-time Algorithm Research Articles

123VCF: an intuitive and efficient tool for filtering VCF files

BackgroundThe advent of Next-Generation Sequencing (NGS) has catalyzed a paradigm shift in medical genetics, enabling the identification of disease-associated variants. However, the vast quantum of data produced by NGS necessitates a robust and dependable mechanism for filtering irrelevant variants. Annotation-based variant filtering, a pivotal step in this process, demands a profound understanding of the case-specific conditions and the relevant annotation instruments. To tackle this complex task, we sought to design an accessible, efficient and more importantly easy to understand variant filtering tool.ResultsOur efforts culminated in the creation of 123VCF, a tool capable of processing both compressed and uncompressed Variant Calling Format (VCF) files. Built on a Java framework, the tool employs a disk-streaming real-time filtering algorithm, allowing it to manage sizable variant files on conventional desktop computers. 123VCF filters input variants in accordance with a predefined filter sequence applied to the input variants. Users are provided the flexibility to define various filtering parameters, such as quality, coverage depth, and variant frequency within the populations. Additionally, 123VCF accommodates user-defined filters tailored to specific case requirements, affording users enhanced control over the filtering process. We evaluated the performance of 123VCF by analyzing different types of variant files and comparing its runtimes to the most similar algorithms like BCFtools filter and GATK VariantFiltration. The results indicated that 123VCF performs relatively well. The tool's intuitive interface and potential for reproducibility make it a valuable asset for both researchers and clinicians.ConclusionThe 123VCF filtering tool provides an effective, dependable approach for filtering variants in both research and clinical settings. As an open-source tool available at https://project123vcf.sourceforge.io, it is accessible to the global scientific and clinical community, paving the way for the discovery of disease-causing variants and facilitating the advancement of personalized medicine.

Research on Indoor 3D Reconstruction Technology Based on Semantic Visual Simultaneous Localization and Mapping

In response to the challenge that traditional visual simultaneous localization and mapping (SLAM) systems, based on the assumption of a static environment, struggle to achieve real-time indoor 3D reconstruction in complex dynamic scenes, this paper proposes a real-time indoor 3D reconstruction algorithm based on semantic visual SLAM. By leveraging object detection to obtain 2D semantic information and providing prior information for geometric methods, the fusion of the two effectively suppresses dynamic features, reduces reliance on deep learning methods, and ensures the algorithm's real-time performance. Experimental results on dynamic scenes in the TUM RGB-D dataset show that our algorithm maintains nearly unchanged real-time performance while achieving an average performance improvement of approximately 97.56% and 97.31% on the TUM dataset and Bonn dataset, respectively, compared to the ORB-SLAM2 system. Moreover, our algorithm can reconstruct more intuitive indoor global Octo-map and semantic metric maps compared to sparse point cloud maps, effectively enhancing the scene perception capability of mobile robots and laying the foundation for performing advanced tasks. Furthermore, our algorithm demonstrates a 3.5-10.5 times improvement in real-time performance compared to other mainstream semantic SLAM systems. Experimental results on the NVIDIA Jetson AGX Xavier confirm that our algorithm can run in real time on low-power platforms such as mobile robots or drones. However, the drawbacks of our algorithm include lower reconstruction accuracy in low-texture and large-scale scenes and ineffective suppression of dynamic features in low-dynamic scenes. Future work will consider replacing and improving deep learning methods and integrating IMU and other sensors to enhance system usability.

Real-time multi-target tracking algorithm based on virtual frame and fuzzy data association

Real-time multi-target tracking algorithm based on virtual frame and fuzzy data association

Enhancing fault tolerance: dual Q-learning with dynamic scheduling

<span>Cloud computing has revolutionized IT delivery by offering scalable on-demand internet services encompassing software, platforms, and infrastructure. However, cloud services face significant performance challenges due to their susceptibility to failures given their vast operational scale. Implementing fault tolerance in dynamic cloud services is a key challenge, with complex configurations and dependencies complicating deployment. This paper introduces an innovative approach that combines double deep Q-learning (DDQL) with a dynamic fault-tolerant real-time scheduling algorithm (DFTRTSA) to enhance fault tolerance in real-time systems. DDQL, an extension of deep Q-learning, optimizes the fault-tolerance decision-making process. The algorithm adjusts scheduling strategies dynamically based on system conditions and errors. The fusion of DDQL and DFTRTSA aims to create a resilient and adaptive fault-tolerant mechanism, ensuring uninterrupted operation while meeting real-time requirements. This adaptive approach efficiently manages resources, meets deadlines, and gracefully handles errors, as demonstrated through experiments. Our DDQL-DFTRTSA method outperforms conventional fault-tolerant mechanisms in defect tolerance, energy efficiency, downtime reduction, and system dependability. It proves to be an ideal solution for real-time systems in dynamic and unpredictable environments.</span>

An efficient lightweight railway track segmentation network for resource-constrained platforms with TensorRT

Abstract Accurate and rapid railway track segmentation is fundamental for foreign object intrusion detection, inspection, online monitoring, and non-destructive assessment of transportation infrastructure. Recently, vision-based track segmentation algorithms have demonstrated strong performance. However, most existing models struggle to meet the real-time requirements of resource-constrained edge devices. Considering this challenge, we propose an edge-enabled real-time railway track segmentation algorithm, which is optimized to be suitable for edge applications by optimizing the network structure and quantizing the model after training. Initially, Ghost convolution is introduced to reduce the complexity of the backbone, thereby achieving the extraction of key information on the region of interest at a lower cost. To further reduce the model complexity and calculation, a new lightweight detection head is proposed to achieve the best balance between accuracy and efficiency. Subsequently, we introduce quantization techniques to map the model’s floating-point weights and activation values into lower bit-width fixed-point representations, reducing computational demands and memory footprint, ultimately accelerating the model’s inference. Finally, we draw inspiration from GPU parallel programming principles to expedite the pre-processing and post-processing stages of the algorithm by doing parallel processing. The approach is evaluated with the public and challenging dataset RailSem19 and tested on Jetson Nano. Experimental results demonstrate that our enhanced algorithm achieves an accuracy level of 83.3% with 25 FPS inference speed when the input size is 480$\times $480. The code can be found at: https://github.com/ccl-1/light-yolov8-seg-quantization-tensorrt.

Stability improvement of 40Ca+ optical clock by using a transportable ultra-stable cavity.

The instability of the clock laser is one of the primary factors limiting the instability of the optical clocks. We present an ultra-stable clock laser based on a 30-cm-long transportable cavity with an instability of ∼3 × 10-16 at 1 s-100 s. The cavity is fixed by invar poles in three orthogonal directions to restrict the displacement, meeting the requirements of transportability and low vibration sensitivity. By applying the ultra-stable laser to a transportable 40Ca+ optical clock with a systematic uncertainty of 4.8 × 10-18 and using the real-time feedback algorithm to compensate the linear shift of the clock laser, the short-term stability of the transportable 40Ca+ optical clock has been greatly improved from 4.0×10-15/τ/s to 1.16×10-15/τ/s, measured at ∼100 s-1000 s of averaging time, enriching its applications in metrology, optical frequency comparison, and time keeping.

A real-time automatic rail extraction algorithm for low-density mobile laser scanning data

Automatic detection of railroad infrastructure using Mobile Laser Scanning systems is a key technology for both advanced rail driver assistance and intelligent track maintenance. The recent research into railway facility extraction or condition monitoring usually relies on high-density point cloud dataset with known sensor parameters but ignores the processing performance in actual deployment and has high requirements for the acquisition device. To address these limitations, a novel rail extraction algorithm is proposed for processing Mobile Laser Scanning data in real time during acquisition, which could extract rail features by a hierarchical coarse-to-fine method with basic structural parameters. Using the geometric and global statistical characteristics of rail in raw data, a new rail descriptor is defined based on the quantitative statistics of points satisfying height difference in generalized local neighborhood. The approach is evaluated experimentally by a simulated real-time acquisition data and compared with a reference method. The experimental results show that the proposed algorithm has a finer extraction effect and good real-time performance.

BDIS-SLAM: a lightweight CPU-based dense stereo SLAM for surgery.

Common dense stereo simultaneous localization and mapping (SLAM) approaches in minimally invasive surgery (MIS) require high-end parallel computational resources for real-time implementation. Yet, it is not always feasible since the computational resources should be allocated to other tasks like segmentation, detection, and tracking. To solve the problem of limited parallel computational power, this research aims at a lightweight dense stereo SLAM system that works on a single-core CPU and achieves real-time performance (more than 30Hz in typical scenarios). A new dense stereo mapping module is integrated with the ORB-SLAM2 system and named BDIS-SLAM. Our new dense stereo mapping module includes stereo matching and 3D dense depth mosaic methods. Stereo matching is achieved with the recently proposed CPU-level real-time matching algorithm Bayesian Dense Inverse Searching (BDIS). A BDIS-based shape recovery and a depth mosaic strategy are integrated as a new thread and coupled with the backbone ORB-SLAM2 system for real-time stereo shape recovery. Experiments on in vivo data sets show that BDIS-SLAM runs at over 30Hz speed on modern single-core CPU in typical endoscopy/colonoscopy scenarios. BDIS-SLAM only consumes around an additional time compared with the backbone ORB-SLAM2. Although our lightweight BDIS-SLAM simplifies the process by ignoring deformation and fusion procedures, it can provide a usable dense mapping for modern MIS on computationally constrained devices. The proposed BDIS-SLAM is a lightweight stereo dense SLAM system for MIS. It achieves 30Hz on a modern single-core CPU in typical endoscopy/colonoscopy scenarios (image size around ). BDIS-SLAM provides a low-cost solution for dense mapping in MIS and has the potential to be applied in surgical robots and AR systems. Code is available at https://github.com/JingweiSong/BDIS-SLAM .

Dense-YOLOv7: improved real-time insulator detection framework based on YOLOv7

Abstract Due to various mechanical and electrical stresses caused by changes in environmental and power load conditions, insulators may fail, resulting in serious economic losses. Manual cleaning of insulators is subject to weather and environmental constraints and poses significant safety risks. Traditional insulator detection methods have problems such as slow detection speed and poor robustness. Therefore, this paper proposes a real-time insulator detection algorithm based on the improved YOLOv7. First, in order to effectively streamline the number of parameters, Dense-YOLOv7 adopts the dense connection concept of DenseNet to design DenseBlock. Second, replacing the loss function with Focal Loss to solve the problem of unbalanced matching of foreground and background sample quantities has improved the detection accuracy of the model. Finally, to address the issue of PReLU and LeakyReLU activation functions being insensitive to spatial information, the activation function is replaced with FReLU to improve the robustness of the model. The experimental dataset used in this paper is a combination of the Chinese Power Transmission Line Insulator Dataset (CPTLID) and our own dataset. The experimental results show that the improved algorithm in this paper has only 44.23 M parameters and a detection speed of 44.87 FPS, which is 4.8% less than that of YOLOv7 and 8.14% quicker than that of YOLOv7. Experimental results show that Dense-YOLOv7 can significantly streamline the model parameter size while maintaining high accuracy and effectively improve the detection speed, which can meet the application requirements of real-time insulator detection.

Electrolytic capacitor surface defect detection based on deep convolution neural network

The existing methods for detecting surface defects in electrolytic capacitors are typically based on conventional machine vision, with limited feature extraction capabilities, poor versatility, slow detection speed, and the inability to achieve accurate and real-time defect detection. In this study, a real-time object detection algorithm based on an improved single shot multibox detector (SSD) is proposed to achieve omnidirectional surface defect detection of electrolytic capacitors. First, an electrolytic capacitor surface image acquisition device was established to capture omnidirectional surface images of the capacitors, and an electrolytic capacitor surface defect dataset was created. Next, the visual geometry group (VGG)-16 network structure was replaced with the MobileNetv2 network structure, effectively reducing the model’s parameter count and improving inference speed. Moreover, the Multibox Loss function was replaced with the Focal Loss function to increase the model’s attention to difficult-to-classify samples and improve model accuracy. Additionally, a transfer learning network model was designed to apply the model to electrolytic capacitors of different colors using small sample learning. Finally, the performance of the improved network model was tested on a dataset of electrolytic capacitor surface defects. The experimental results demonstrate that the parameters quantity of improved model is 3.50 M, the mAP value reaches 92.67 %, which is improved by 2.54 %, and the Macro-F1 value reaches 92.15 %, which is 11.32 % higher than that before improvement. Thus, the proposed improved SSD model provides a theoretical basis and technical prerequisites for automated and intelligent surface defect detection in electrolytic capacitors.

METS-R SIM: A simulator for Multi-modal Energy-optimal Trip Scheduling in Real-time with shared autonomous electric vehicles

We develop an agent-based simulator named METS-R SIM to support operational decisions for multi-modal shared autonomous vehicle (SAEV) services. Compared with existing traffic simulators, METS-R SIM offers several valuable features including: 1) A microscopic vehicle movement model for SAEV services, which allows us to explicitly model vehicular interactions and generate detailed speed and acceleration profiles for energy estimation. 2) An efficient implementation in which parallel computing is embedded in METS-R SIM which can update the state of different agents (e.g., vehicle locations in different links) simultaneously. 3) A modular and extensible framework as the simulator is built upon an agent-based modeling environment named Repast Simphony which is featured by its well-factored abstractions; in addition, a server-client structure is introduced to implement real-time operational algorithms such as energy-efficient routing and adaptive transit scheduling. 4) Open-source, reproducible with web-based visualization (METS-R SIM introduces these features to promote transparency). We validate METS-R SIM by matching the aggregated travel time and travel distance with the real observed ones obtained from New York City (NYC). We also compare the generated speed profiles qualitatively to the ones reported in published studies. We demonstrate the functionalities of our simulator by simulating SAEV services deployed to serve travel needs related to three main transportation hubs in NYC.

Psychological experiences and travel Adversities: A Mixed-Method study of the regular commuters in traffic congestion

This study investigated the psychological experiences and consequences of travel adversities during traffic congestion using a three-phase sequential exploratory mixed-methods design. Phase 1 explored the travel adversities, psychological experiences, and consequences of a sample of ten (four women and six men) regular commuters of Bangalore's congested roads using semi-structured interviews. In phase 2, a checklist was developed listing the fundamental themes from phase 1 with Likert-type responses ranging from 0 (never) to 5 (always). Phase 3 gathered data in the checklist and tested the statistical validity of the thematic model in a sample of 190 (81 women and 103 men) regular commuters. Attride-Stirling model thematic network was established with 57 fundamental themes categorized and assigned under the organizing themes of travel adversities (n = 6), negative affect (n = 28), fight (n = 7), flight (n = 6), and negative road occurrences (n = 10), in the global theme, psychological experiences and consequences. Structural equation modeling indicated that (1) negative affect significantly predicted fight and flight, (2) fight is a significant predictor of negative road occurrences, and (3) psychological experiences and consequences create a self-perpetuating cycle, with travel adversity triggering negative emotions, which results in fight responses leading to negative road occurrences, further intensifying travel adversity. A mathematical model is established based on this statistical validation, which holds potential applications in real-time traffic algorithms.

CyberWatch: Deep Learning-Driven Network Intrusion Detection

Abstract—The present era is being dominated by the digital world which has also given rise to growing cyber threats and crimes, the project introduces an innovative intrusion detection system(IDS) that deploys deep learning’s pattern recognition capabilities with a real-time algorithm approach to threats and network security. The project’s framework involves two key components: Graph theory and Artificial Neural Network (ANN). The graph theory is used to represent the IoT network structure which helps represent relations between network structures and analyze the anomalies or malicious activities present in the network. Artificial Neural Network (ANN) model, on the other hand, is based on a human’s brain neural network; which works as a machine learning algorithm to recognize and predict patterns. Cyberwatch is trained on various datasets such as CIC IDS 2018 and CIC IDS 2017. The project’s main aim is to provide an effective solution to the ever-changing landscape of network intrusion. Keywords—Intrusion Detection System(IDS), Energy Ef- ficiency, Internet Of Things(IoT), Network Forensics, Graph Theory Support Vector Machine(SVM), Genetic Algorithm, Artificial Neural Networks(ANN), Cybercrime

TLW: A Real-Time Light Curve Classification Algorithm for Transients Based on Machine Learning

The real-time light curve classification of transients is helpful in searching for rare transients. We propose a new algorithm based on machine learning, namely the Temporary Convective Network and Light Gradient Boosting Machine Combined with Weight Module Algorithm (TLW). The TLW algorithm can classify the photometric simulation transients data in g, r, i bands provided via PLAsTiCC, typing Tidal Disruption Event (TDE), Kilonova (KN), Type Ia supernova (SNIa), and Type I Super-luminous supernova (SLSN-I). When comparing the real-time classification results of the TLW algorithm and six other algorithms, such as Rapid, we found that the TLW algorithm has the best comprehensive performance indexes and has the advantages of high precision and high efficiency. The average accuracy of TLW is 84.54%. The average implementation timings of the TLW algorithm for classifying four types of transients is 123.09 s, which is based on TensorFlow’s architecture in windows and python. We use three indicators to prove that the TLW algorithm is superior to the classical Rapid algorithm, including Confusion Matrix, PR curve, and ROC curve. We also use the TLW algorithm to classify ZTF real transients. The real-time classification results for ZTF transients show that the accuracy of the TLW algorithm is higher than the other six algorithms.

Deep learning and multi-modal fusion for real-time multi-object tracking: Algorithms, challenges, datasets, and comparative study

Real-time multi-object tracking (MOT) is a complex task involving detecting and tracking multiple objects. After the objects are detected, they are assigned markers, and their trajectories are tracked in real-time. The scientific community is intrigued by the possibilities of utilizing MOT technology in the context of smart cities. Their primary focus lies in the domains of intelligent transportation, detection of vehicles and pedestrians, crowd surveillance, and public safety. Deep learning techniques have been developed in recent years to effectively tackle the challenges of real-time MOT tasks and enhance tracking performance. Environmental perception within smart traffic applications relies heavily on sensor data fusion. In traffic scenarios, a thoughtful approach involves utilizing a combination of sensors and cameras to detect and track targets while gathering valuable data effectively. However, it faces challenges when it comes to detecting and tracking objects that are in motion, have complex changes in appearance, or are in crowded scenes. This paper explores the foundational standard for real-time Multiple Object Tracking tasks. We prioritize the examination of quantitative measures by conducting a comprehensive analysis of widely utilized benchmark datasets and metrics. This study also investigates established embedding techniques and multi-modal fusion methods within real-time multi-target tracking algorithms. Each strategy will be classified and assessed according to a predefined set of principles. The paper presents a comprehensive analysis and visual representation of various MOT strategies. Finally, this paper aims to present an overview of the current challenges faced by the MOT mission, as well as the potential objectives that lie ahead.

Application of a real-time flame smoke detection algorithm based on improved YOLOv7

Flame and smoke detection is a critical issue that has been widely used in various unmanned security monitoring scenarios. However, existing flame smoke detection methods suffer from low accuracy and slow speed, and these problems reduce the efficiency of real-time detection. To solve the above problems, we propose an improved YOLOv7(You Only Look Once) algorithm for flame smoke mobile detection. The algorithm uses the Kmeans algorithm to cluster the prior frames in the dataset and uses a lightweight CNeB(ConvNext Block) module to replace part of the traditional ELAN module to accelerate the detection speed while ensuring high accuracy. In addition, we propose an improved CIoU loss function to further enhance the detection effect. The experimental results show that, compared with the original algorithm, our algorithm improves the accuracy by 4.5% and the speed by 39.87%. This indicates that our algorithm meets the real-time monitoring requirements and can be practically applied to field detection on mobile edge computing devices.

Adaptive Intervention Algorithms for Advanced Driver Assistance Systems

Advanced driver assistance systems (ADASs) have recently gained popularity as they assist vehicle operators in staying within safe boundaries, helping them thereby to prevent possible collisions. However, despite their success and development, most ADAS use common and deterministic warning thresholds for all drivers in all driving environments. This may occasionally lead to the issuance of annoying inadequate warnings, due to the possible differences between drivers, the changing environments and driver statuses, thus reducing their acceptance and effectiveness. To fill this gap, this paper proposes adaptive algorithms for commonly used warnings based on real-time traffic environments and driver status including distraction and fatigue. We proposed adaptive algorithms for headway monitoring, illegal overtaking, over-speeding, and fatigue. The algorithms were then tested using a driving simulator. Results showed that the proposed adaptive headway warning algorithm was able to automatically update the headway warning thresholds and that, overall, the proposed algorithms provided the expected warnings. In particular, three or four different warning types were designed to distinguish different risk levels. The designed real-time intervention algorithms can be implemented in ADAS to provide warnings and interventions tailored to the driver status to further ensure driving safety.

Real-time ghost imaging algorithm on the multidimensional vector matrix Walsh transformation with spatiotemporal free-fps.

In this research, we propose a real-time spatiotemporal free-fps algorithm based on the multidimensional vector matrix Walsh transform with an adjustable ghost imaging video quality and a frame rate to address the issues of uneven imaging speed, fixed imaging frame rate, and uncomfortable appearance in real-time ghost imaging videos of moving objects. This algorithm utilizes the temporal and spatial correlation of ghost imaging videos to achieve free and adjustable video frame rates in time without being limited by DMD refresh rates. Improving the spatial information of a single frame in space enhances the smoothness of ghost imaging videos, making the appearance of ghost imaging videos more comfortable. To achieve this, a four-dimensional vector Walsh transform kernel matrix is used to transform and reconstruct the high-quality images of the target object. Then the reconstructed high-quality image is spatially interpolated to enhance spatial information. Reasonable frame rate parameters are set based on the corresponding relationship between the detection values of the adjacent frames and speckle, improving the ghost imaging video in both time and space and achieving a smooth real-time ghost imaging video with an adjustable quality and frame rate. The simulation and experimental results of moving objects show that our algorithm solves the limitation of a DMD refresh rate compared with the existing ghost imaging video methods and makes the ghost imaging video more comfortable and smoother in real time. The PSNR of the objective evaluation index is increased by 12%. Regarding a subjective evaluation, this paper proposes an adaptive parameterless evaluation algorithm (APEA) for images with different resolutions based on the NRSS, which improved the structure retention degree by 13% and the Brisque parameter evaluation by 70%. We propose an adaptive parameterless video comfort evaluation algorithm (APVCEA) to evaluate the subjective comfort of ghost imaging videos by 14% compared with traditional methods.

DLD-SLAM: RGB-D Visual Simultaneous Localisation and Mapping in Indoor Dynamic Environments Based on Deep Learning

This work presents a novel RGB-D dynamic Simultaneous Localisation and Mapping (SLAM) method that improves the precision, stability, and efficiency of localisation while relying on lightweight deep learning in a dynamic environment compared to the traditional static feature-based visual SLAM algorithm. Based on ORB-SLAM3, the GCNv2-tiny network instead of the ORB method, improves the reliability of feature extraction and matching and the accuracy of position estimation; then, the semantic segmentation thread employs the lightweight YOLOv5s object detection algorithm based on the GSConv network combined with a depth image to determine potentially dynamic regions of the image. Finally, to guarantee that the static feature points are used for position estimation, dynamic probability is employed to determine the true dynamic feature points based on the optical flow, semantic labels, and the state in last frame. We have performed experiments on the TUM datasets to verify the feasibility of the algorithm. Compared with the classical dynamic visual SLAM algorithm, the experimental results demonstrate that the absolute trajectory error is greatly reduced in dynamic environments, and that the computing efficiency is improved by 31.54% compared with the real-time dynamic visual SLAM algorithm with close accuracy, demonstrating the superiority of DLD-SLAM in accuracy, stability, and efficiency.

An efficient segmentation model for abnormal chicken droppings recognition based on improved deep dual-resolution network.

The characteristics of chicken droppings are closely linked to their health status. In prior studies, chicken droppings recognition is treated as an object detection task, leading to challenges in labeling and missed detection due to the diverse shapes, overlapping boundaries, and dense distribution of chicken droppings. Additionally, the use of intelligent monitoring equipment equipped with edge devices in farms can significantly reduce manual labor. However, the limited computational power of edge devices presents challenges in deploying real-time segmentation algorithms for field applications. Therefore, this study redefines the task as a segmentation task, with the main objective being the development of a lightweight segmentation model for the automated monitoring of abnormal chicken droppings. A total of 60 Arbor Acres broilers were housed in 5 specific pathogen-free cages for over 3 weeks, and 1650 RGB images of chicken droppings were randomly divided into training and testing sets in an 8:2 ratio to develop and test the model. Firstly, by incorporating the attention mechanism, multi-loss function and auxiliary segmentation head, the segmentation accuracy of the DDRNet was enhanced. Then, by employing the group convolution and an advanced knowledge distillation algorithm, a lightweight segmentation model named DDRNet-s-KD was obtained, which achieved a mean Dice coefficient (mDice) of 79.43% and an inference speed of 86.10 frames per second (FPS), showing a 2.91% and 61.2% increase in mDice and FPS compared to the benchmark model. Furthermore, the DDRNet-s-KD model was quantized from 32-bit floating-point values to 8-bit integers and then converted to TensorRT format. Impressively, the weight size of the quantized model was only 13.7 MB, representing an 82.96% reduction compared to the benchmark model. This makes it well-suited for deployment on the edge device, achieving an inference speed of 137.51 FPS on Jetson Xavier NX. In conclusion, the methods proposed in this study show significant potential in monitoring abnormal chicken droppings and can provide an effective reference for the implementation of other agricultural embedded systems.