Original Algorithm Research Articles

Tracking and automatic behavioral analysis of group-housed pigs based on YOLOX+BoT-SORT-slim

Automatic analysis of pig behavior is crucial for assessing their health and welfare status in group-housed pig farms. Currently, computer vision technology has been widely applied to recognize and track pig behavior. However, these applications lack the capability for automatic behavior analysis. Additionally, execution speed is crucial for long-term tracking, as the tracking target can easily be lost due to the complex background of pig farms, uneven lighting, and the similar appearance of pigs. To tackle these challenges, we propose a lightweight multi-object tracking (MOT) and behavioral analysis method: YOLOX+BoT-SORT-Slim. Firstly, the YOLOX detector detects pig targets and recognizes the pigs’ four behaviors including “stand”, “lie”, “eat”, and “other”. Secondly, the BoT-SORT-Slim algorithm tracks the detected pigs and their behaviors at high speed. Finally, we devise a behavior automatic analysis algorithm that integrates the behavioral information and the tracking results to complete pigs’ behavior analysis. To evaluate the proposed method, we conducted experiments on 1-minute, 10-minute, and 1-hour datasets. The experimental results show that on the 1-minute public and private test sets, the proposed method achieves 6.4 and 3.6 times the frame rate (FPS) of the original algorithm, with 68.05 FPS and 71.27 FPS, respectively. It also achieves a higher-order tracking accuracy (HOTA) of 84.7 % and 78.7 %, multi-object tracking accuracy (MOTA) of 98.3 % and 97.0 %, and an identification F1 score (IDF1) of 98.4 % and 93.4 %. The proposed method obtains the best results on all test sets compared to five other MOT algorithms including DeepSORT, StrongSORT, ByteTrack, OC-SORT, and BoT-SORT. In the 10-minute and 1-hour test sets, the proposed method achieves 70.28 and 72.24, 63.3 % and 76.4 %, 94.1 % and 99.6 %, and 72.5 % and 90.2 % for FPS, HOTA, MOTA, and IDF1, respectively. In terms of behavior analysis, three types of pigs’ behavior analysis are plotted for each test video to assess their health status. The experimental results indicate that the proposed method performs excellently in behavior analysis and tracking, which can reliably monitor and manage pig behavior in real time, providing automated management technical support.

Fault detection method for transmission line components based on lightweight GMPPD-YOLO

Abstract This paper designs a lightweight high-precision transmission line component detection model, named grouped dense, monotonic self-regularized, and partial faster convolution, pruning, and distillation optimized—you only look once (GMPPD-YOLO), in transmission line inspection. It addresses the issue of low detection accuracy of target detection algorithms due to the complex background, large differences in target shape, location, texture, etc, as well as diversified and smaller defects in insulator and vibration hammer images taken by unmanned aerial vehicles from multiple angles. To enhance the model’s feature extraction capabilities in complex backgrounds and across different scales, the grouped dense C3 dense feature extraction module was designed, enabling the model to more effectively handle diverse defect forms. Simultaneously, the monotonic self-regularized pyramid pooling–fast (MSPPF) module is proposed to enhance the model’s capability to process multi-scale information. Additionally, the partial-faster C3 feature awareness module is designed to improve feature fusion performance, enhancing the model’s ability to perceive features at different scales. Finally, channel pruning was used to reduce redundant parameters, and knowledge distillation was employed to compensate for the accuracy loss caused by pruning. This approach further compressed the model size while ensuring its detection performance. The experimental results demonstrate that compared to the original YOLOv5s algorithm, the proposed GMPPD-YOLO algorithm achieves a reduction in parameters by 68.4%, a decrease in Giga floating-point operations per second by 58.2%, and a reduction in the model size by 66.4%, while achieving an increase in precision by 1%, mAP50 by 1.1%, and mAP95 by 0.4%. This confirms the significant potential of the GMPPD-YOLO algorithm for deployment in real-time drone-based power transmission line inspections.

A Method for Recognition and Coordinate Reference of Autonomous Underwater Vehicles to Inspected Objects of Industrial Subsea Structures Using Stereo Images

To date, the development of unmanned technologies using autonomous underwater vehicles (AUVs) has become an urgent demand for solving the problem of inspecting industrial subsea structures. A key issue here is the precise localization of AUVs relative to underwater objects. However, the impossibility of using GPS and the presence of various interferences associated with the dynamics of the underwater environment do not allow high-precision navigation based solely on a standard suite of AUV navigation tools (sonars, etc.). An alternative technology involves the processing of optical images that, at short distances, can provide higher accuracy of AUV navigation compared to the technology of acoustic measurement processing. Although there have been results in this direction, further development of methods for extracting spatial information about objects from images recorded by a camera is necessary in the task of calculating the exact mutual position of the AUV and the object. In this study, in the context of the problem of subsea production system inspection, we propose a technology to recognize underwater objects and provide coordinate references to the AUV based on stereo-image processing. Its distinctive features are the use of a non-standard technique to generate a geometric model of an object from its views (foreshortening) taken from positions of a pre-made overview trajectory, the use of various characteristic geometric elements when recognizing objects, and the original algorithms for comparing visual data of the inspection trajectory with an a priori model of the object. The results of experiments on virtual scenes and with real data showed the effectiveness of the proposed technology.

Infrared Weak and Small Target Detection Algorithm Based on Deep Learning

Abstract In the infrared imaging scene where the target is at a long distance and the background is cluttered, due to the interference of noise and background texture information, the infrared image is prone to problems such as low contrast between the target and the background, and feature confusion, which makes it difficult to accurately extract and detect the target. To solve this problem, firstly, the infrared image is enhanced by combining DDE and MSR algorithm to improve the contrast and detail visibility of the image. For the RT-DETR network structure, the EMA attention mechanism is introduced into the backbone to enhance the feature extraction ability of the model by extracting context information. The CAMixing convolutional attention module is introduced into CCFM, and the multi-scale convolutional self-attention mechanism is introduced to focus on local information and enhance the detection ability of small targets. The filtering rules of the prediction box are improved, combined with Shape-IoU, and the convergence speed of the loss function in the detection and the detection accuracy of small targets are improved by paying attention to the influence of the intrinsic properties of the bounding box itself on the regression. In the experiment, the infrared weak target image dataset of the National University of Defense Technology was selected, labeled and trained. Experimental results show that compared with the original DETR algorithm, the average precision of the improved algorithm (mAP) is increased by 3.2%, and it can effectively detect infrared weak and small targets in different complex backgrounds, which reflects good robustness and adaptability, and can be effectively applied to infrared weak and small target detection in complex backgrounds.

Improved Pedestrian Vehicle Detection for Small Objects Based on Attention Mechanism

Abstract This study aims to solve the low detection accuracy and susceptibility to false detection and omission in pedestrian and vehicle detection by proposing an improved YOLOv5s algorithm. Firstly, a small target detection module is added to better acquire and determine the information of pedestrians from long-range vehicles. Secondly, the multi-scale channel attention CBAM attention module is added, and the dual attention mechanism is not only flexible and convenient, but also improves the computational efficiency. Finally, the MPDIoU loss function based on minimum point distance is introduced to replace the original GIoU loss function, and this change not only enhances the regression accuracy of the model. At the same time, the convergence speed of the model is accelerated. KITTI data set was used for experiments, and the experimental results showed that the average accuracy of the model trained by the improved YOLOv5s algorithm on the data set reached 84.9%, which was 3.7% higher than that of the original YOLOv5s algorithm. It is verified that the model is suitable for high accuracy of pedestrian and vehicle recognition in complex environments, and has high value for promotion.

Insulator defect detection algorithm based on YOLOv8-ISDD

Abstract Insulators are widely used in transmission lines due to their good insulation performance and pollution resistance. The damage to insulators will affect their service life and insulation performance, which brings potential safety hazards to transmission lines. With the replacement of manual inspection with UAV inspection, image detection technology based on deep learning is gradually combined with UAV inspection. Usually, the insulator defect target accounts for a small proportion of the acquired image. Therefore, the manuscript introduces an enhanced algorithm for detecting flaws on the surface of insulators, designated as YOLOv8-ISDD. In the backbone network, the YOLOv8-ISDD algorithm introduces reparameterized convolution (Repconv) to enhance the network’s capability in extracting features of defects; to fuse the features of different resolutions, PANet in the neck network is replaced by Bi-FPN for integrating features. Within the head network, the multi-scale detection head output is used to increase the diminutive object identification stratum and boost the algorithm’s capability to discern lesser targets. Experiments show that the mAP@0.5 of the YOLOv8-ISDD algorithm is 88.9%, which is 6% higher than that of the original YOLOv8 algorithm, indicating that YOLOv8-ISDD can detect insulator defects more accurately.

Improved LPAstar algorithm for power station cabinet wiring

Abstract Aiming at the complex constraints of cable layout in power plant cabinets and the low efficiency of manual design, an improved Lifelong Planning Astar algorithm (LPAstar) is proposed. By constructing a corner-length feature fusion model, the cable length and number of corners are constrained to reduce wiring costs and cable damage. The spatial minimum bounding box algorithm is employed to avoid cable suspension. Continuous incentive factors are introduced to improve algorithm efficiency. In post-processing, smoothing strategy and wire harness dispersion strategy are used to avoid excessive cable bending and ensure the standardization of wiring. The simulation experiment results show that compared with the traditional Astar algorithm, the routing length and number of corners are reduced by 27% and 65%, respectively. The generation efficiency is 84% and 87% higher than the Astar algorithm and the original LPAstar algorithm, respectively.

Paper-Recorded ECG Digitization Method with Automatic Reference Voltage Selection for Telemonitoring and Diagnosis.

In electrocardiograms (ECGs), multiple forms of encryption and preservation formats create difficulties for data sharing and retrospective disease analysis. Additionally, photography and storage using mobile devices are convenient, but the images acquired contain different noise interferences. To address this problem, a suite of novel methodologies was proposed for converting paper-recorded ECGs into digital data. Firstly, this study ingeniously removed gridlines by utilizing the Hue Saturation Value (HSV) spatial properties of ECGs. Moreover, this study introduced an innovative adaptive local thresholding method with high robustness for foreground-background separation. Subsequently, an algorithm for the automatic recognition of calibration square waves was proposed to ensure consistency in amplitude, rather than solely in shape, for digital signals. The original signal reconstruction algorithm was validated with the MIT-BIH and PTB databases by comparing the difference between the reconstructed and the original signals. Moreover, the mean of the Pearson correlation coefficient was 0.97 and 0.98, respectively, while the mean absolute errors were 0.324 and 0.241, respectively. The method proposed in this study converts paper-recorded ECGs into a digital format, enabling direct analysis using software. Automated techniques for acquiring and restoring ECG reference voltages enhance the reconstruction accuracy. This innovative approach facilitates data storage, medical communication, and remote ECG analysis, and minimizes errors in remote diagnosis.

Lithium battery state of charge estimation based on improved variable forgetting factor recursive least squares method and adaptive Kalman filter joint algorithm

As one of the core functions of the battery management system, battery state of charge (SOC) estimation is crucial to battery life and safety. As the traditional recursive least squares method cannot adapt well to complex variable current conditions, this paper proposes a SOC estimation method that combines the improved variable forgetting factor recursive least squares (IVFFRLS) and the adaptive extended Kalman filter (AEKF). First, IVFFRLS removes the rounding operation of the original algorithm to improve the sensitivity to errors and uses a genetic algorithm to optimize the upper and lower bounds of the variable forgetting factor and the sensitivity coefficient to improve the accuracy of the VFFRLS algorithm. Meanwhile, AEKF introduces a noise covariance adaptive update link based on the covariance matching method to correct the system's process noise Q and measurement noise R in real time to improve the SOC estimation accuracy. Based on the second-order RC equivalent circuit model, the method proposed in this paper has good accuracy and robustness under four different working conditions. The experimental results show that compared with traditional methods, the proposed IVFFRLS-AEKF algorithm has higher SOC estimation accuracy and better robustness.

Lightweight Algorithm for Rail Fastener Status Detection Based on YOLOv8n

To improve the accuracy of rail fastener detection and deploy deep learning models on mobile platforms for fast real-time inference, this paper proposes a defect detection model for rail fasteners based on an improved YOLOv8n. Considering the significant aspect ratio differences of rail fasteners, we designed the EIOU+ as the regression box loss function. The model is compressed and trained using an improved channel-wise knowledge distillation (CWD+) approach to address the challenge of accurately recognizing minor defects in rail fasteners. We introduced a feature extraction module to design a feature extraction network as the distillation teacher model (YOLOv8n-T) and a lightweight cross-stage partial bottleneck with two convolutions and a fusion module (C2f) to improve the YOLOv8n backbone network as the distillation student model (YOLOv8n-S). Experiments conducted on data collected from actual rail lines demonstrate that after CWD+ distillation training, the model’s mean detection accuracy (IOU = 0.5) reached 96.3%, an improvement of 2.7% over the original YOLOv8n algorithm. The recall rate increased by 4.5%, the precision by 2.7%, the number of floating-point operations decreased by 13%, and the detection frame rate frames per second (FPS) increased by 6.1 frames per second. Compared with other one-stage object detection algorithms, the CWD+ distilled model achieves the precise real-time detection of rail fastener conditions.

Determination of Hydrogen Bonds in GROMACS: A New Implementation to Overcome Memory Limitation.

This work introduces a new faster implementation of the hydrogen bond network (complex arrangement of hydrogen bonds between or within molecules) search algorithm in biomacromolecules and their environment. Existing implementation of such an algorithm in GROMACS [Abraham et al. GROMACS 2024.2 Manual. 2024.] has limitations in the analysis of large structures and trajectories. The new implementation, in the form of a native GROMACS trajectory analysis module, allows for quick analysis of molecular dynamics trajectories without restrictions, thus overcoming the limitations of the original algorithm. The application of the developed method enabled the acquisition and analysis of hydrogen bond networks in the studied defensin-like protein Pentadiplandra brazzeana, as well as the study of hydrogen bond occupancies between the protein's residues and water molecules. The data obtained using the new implementation coincided with the experimental data.

EA‐YOLO: An Efficient and Accurate UAV Image Object Detection Algorithm

An improved EA‐YOLO object detection algorithm based on YOLOv5 is proposed to address the issues of drastic changes in target scale, low detection accuracy, and high miss rate in unmanned aerial vehicle aerial photography scenarios. Firstly, a DFE module was proposed to improve the effectiveness of feature extraction and enhance the whole model's ability to learn residual features. Secondly, a CWFF architecture was introduced to enable deeper feature fusion and improve the effectiveness of feature fusion. Finally, in order to solve the traditional algorithm's shortcomings it is difficult to detect small targets. We have designed a novel SDS structure and adopted a strategy of reusing low‐level feature maps to enhance the network's ability to detect small targets, making it more suitable for detecting some small objects in drone images. Experiments in the VisDrone2019 dataset demonstrated that the proposed EA‐YOLOs achieved an average accuracy mAP@0.5 of 39.9%, which is an 8% improvement over YOLOv5s, and mAP@0.5:0.95 of 22.2%, which is 5.2% improvement over the original algorithm. Compared with YOLOv3, YOLOv5l, and YOLOv8s, the mAP@0.5 of EA‐YOLOs improved by 0.9%, 1.8%, and 0.6%, while the GFLOPs decreased by 86.4%, 80.6%, and 26.7%. © 2024 Institute of Electrical Engineers of Japan and Wiley Periodicals LLC.

A Novel Fusion Perception Algorithm of Tree Branch/Trunk and Apple for Harvesting Robot Based on Improved YOLOv8s

Aiming to accurately identify apple targets and achieve segmentation and the extraction of branch and trunk areas of apple trees, providing visual guidance for a picking robot to actively adjust its posture to avoid branch trunks for obstacle avoidance fruit picking, the spindle-shaped fruit trees, which are widely planted in standard modern apple orchards, were focused on, and an algorithm for apple tree fruit detection and branch segmentation for picking robots was proposed based on an improved YOLOv8s model design. Firstly, image data of spindle-shaped fruit trees in modern apple orchards were collected, and annotations of object detection and pixel-level segmentation were conducted on the data. Training set data were then augmented to improve the generalization performance of the apple detection and branch segmentation algorithm. Secondly, the original YOLOv8s network architecture’s design was improved by embedding the SE module visual attention mechanism after the C2f module of the YOLOv8s Backbone network architecture. Finally, the dynamic snake convolution module was embedded into the Neck structure of the YOLOv8s network architecture to better extract feature information of different apple targets and tree branches. The experimental results showed that the proposed improved algorithm can effectively recognize apple targets in images and segment tree branches and trunks. For apple recognition, the precision was 99.6%, the recall was 96.8%, and the mAP value was 98.3%. The mAP value for branch and trunk segmentation was 81.6%. The proposed improved YOLOv8s algorithm design was compared with the original YOLOv8s, YOLOv8n, and YOLOv5s algorithms for the recognition of apple targets and segmentation of tree branches and trunks on test set images. The experimental results showed that compared with the other three algorithms, the proposed algorithm increased the mAP for apple recognition by 1.5%, 2.3%, and 6%, respectively. The mAP for tree branch and trunk segmentation was increased by 3.7%, 15.4%, and 24.4%, respectively. The proposed detection and segmentation algorithm for apple tree fruits, branches, and trunks is of great significance for ensuring the success rate of robot harvesting, which can provide technical support for the development of an intelligent apple harvesting robot.

Joint optimization of truck-drone routing for last-mile deliveries in urban areas

This paper presents a joint routing optimization model to satisfy the on-demand customer requirements by minimising the transportation and time penalty costs of trucks and drones, considering the constraints on paths and delivery time of trucks and drones, the drone battery capacity limit, and the customer demands. The improved genetic algorithm is proposed to solve the optimization model followed by the computational experiments demonstrating the superiority of the improved genetic algorithm over the original genetic algorithm in computational efficiency, economy, and punctuality. Further, the comparative analysis demonstrates that the multi-drone joint delivery mode considering the customer demands performs better than the other delivery modes and that the delivery routes can be re-optimised promptly to satisfy the real-time customer demands. The sensitivity analysis on the drone design parameters provides theoretical insights into deploying the size and speed of drone platoons when promoting truck-drone joint delivery in applications.

An Enhanced SL-YOLOv8-Based Lightweight Remote Sensing Detection Algorithm for Identifying Broken Strands in Transmission Lines

Power transmission lines frequently face threats from lightning strikes, severe storms, and chemical corrosion, which can lead to damage in steel–aluminum-stranded wires, thereby seriously affecting the stability of the power system. Currently, manual inspections are relatively inefficient and high risk, while drone inspections are often limited by complex environments and obstacles. Existing detection algorithms still face difficulties in identifying broken strands. To address these issues, this paper proposes a new method called SL-YOLOv8. This method incorporates an improved You Only Look Once version 8 (YOLOv8) algorithm, specifically designed for online intelligent inspection robots to detect broken strands in transmission lines. Transmission lines are susceptible to lightning strikes, storms, and chemical corrosion, which is leading to the potential failure of steel- and aluminum-stranded lines, and significantly impacting the stability of the power system. Currently, manual inspections come with relatively low efficiency and high risk, and Unmanned Aerial Vehicle (UAV) inspections are hindered by complex situations and obstacles, with current algorithms making it difficult to detect the broken strand lines. This paper proposes SL-YOLOv8, which is a broken transmission line strand detection method for an online intelligent inspection robot combined with an improved You Only Look Once version 8 (YOLOv8). By incorporating the Squeeze-and-Excitation Network version 2 (SENet_v2) into the feature fusion network, the method effectively enhances adaptive feature representation by focusing on and amplifying key information, thereby improving the network’s capability to detect small objects. Additionally, the introduction of the LSKblockAttention module, which combines Large Selective Kernels (LSKs) and the attention mechanism, allows the model to dynamically select and enhance critical features, significantly enhancing detection accuracy and robustness while maintaining model precision. Compared with the original YOLOv8 algorithm, SL-YOLOv8 demonstrates improved precision recognition accuracy in Break-ID-1632 and cable damage datasets. The precision is increased by 3.9% and 2.7%, and the recall is increased by 12.2% and 2.3%, respectively, for the two datasets. The mean average precision (mAP) at the Intersection over Union (IoU) threshold of 0.5 is also increased by 4.9% and 1.2%, showing the SL-YOLOv8’s effectiveness in accurately identifying small objects in complex situations.

A Small-Object-Detection Algorithm Based on LiDAR Point-Cloud Clustering for Autonomous Vehicles.

3D object-detection based on LiDAR point clouds can help driverless vehicles detect obstacles. However, the existing point-cloud-based object-detection methods are generally ineffective in detecting small objects such as pedestrians and cyclists. Therefore, a small-object-detection algorithm based on clustering is proposed. Firstly, a new segmented ground-point clouds segmentation algorithm is proposed, which filters out the object point clouds according to the heuristic rules and realizes the ground segmentation by multi-region plane-fitting. Then, the small-object point cloud is clustered using an improved DBSCAN clustering algorithm. The K-means++ algorithm for pre-clustering is used, the neighborhood radius is adaptively adjusted according to the distance, and the core point search method of the original algorithm is improved. Finally, the detection of small objects is completed using the directional wraparound box model. After extensive experiments, it was shown that the precision and recall of our proposed ground-segmentation algorithm reached 91.86% and 92.70%, respectively, and the improved DBSCAN clustering algorithm improved the recall of pedestrians and cyclists by 15.89% and 9.50%, respectively. In addition, visualization experiments confirmed that our proposed small-object-detection algorithm based on the point-cloud clustering method can realize the accurate detection of small objects.

Beating one bit of communication with and without quantum pseudo-telepathy

According to Bell’s theorem, certain entangled states cannot be simulated classically using local hidden variables (LHV). Suppose that we can augment LHV by some amount of classical communication. The question then arises as to how many bits are needed to simulate entangled states? There is very strong evidence that a single bit of communication is powerful enough to simulate projective measurements on any two-qubit entangled state. However, the problem of simulating measurements on higher-dimensional systems remains largely unexplored. In this study, we present Bell-like scenarios, even with three inputs per party, in which bipartite correlations resulting from measurements on higher-dimensional states cannot be simulated with a single bit of communication. We consider the case where the communication direction is fixed and the case where it is bidirectional. To this end, we introduce constructions based on parallel repetition of pseudo-telepathy games and an original algorithm based on branch-and-bound technique to compute the one-bit classical bound. Two copies of emblematic Bell expressions, such as the Magic square pseudo-telepathy game, prove to be particularly powerful, requiring a 16 × 16 state to beat the bidirectional one-bit classical bound, and look a promising candidate for implementation on an optical platform.

Dynamic Tracking Method Based on Improved DeepSORT for Electric Vehicle

The development of electric vehicles has facilitated intelligent transportation, which requires the swift and effective detection and tracking of moving vehicles. To satisfy this demand, this paper presents an enhanced DeepSORT algorithm. By selecting YOLO-SSFS as the front-end detector and incorporating a lightweight and high-precision feature training network called FasterNet, the proposed method effectively extracts vehicle appearance attributes. Besides this, the noise scale adaptive Kalman filter is implemented and the conventional cascade matching process is substituted with global join matching, thereby enhancing overall performance and tracking accuracy. Validation conducted on the VisDrone dataset demonstrates the superiority of this method compared to the original DeepSORT algorithm, exhibiting a 4.76% increase in tracking accuracy and a 3.10% improvement in tracking precision. The findings reveal the advantages of the algorithms in the domain of vehicle detection and tracking, allowing significant technological advancements in intelligent transportation systems.

LUD-YOLO: A novel lightweight object detection network for unmanned aerial vehicle

Autonomous execution of tasks by unmanned aerial vehicles (UAVs) relies heavily on object detection. However, object detection in most images presents challenges such as complex backgrounds, small targets, and obstructions. Additionally, the limited computing speed and memory of the UAV processor affects the accuracy of conventional object detection algorithms. This paper proposes LUD-You Only Look Once (YOLO), a small and lightweight object detection algorithm for UAVs based on YOLOv8. The proposed algorithm introduces a new multiscale feature fusion mode that solves the degradation in feature propagation and interaction through the introduction of upsampling in the feature pyramid network and the progressive feature pyramid network. The application of the dynamic sparse attention mechanism in the Cf2 module achieves flexible computing allocation and content awareness. Furthermore, the proposed model is optimized to be sparse and lightweight, making it possible to deploy on UAV edge devices. Finally, the effectiveness and superiority of LUD-YOLO were verified on the VisDrone2019 and UAVDT datasets. The results of ablation and comparison experiments show that compared with the original algorithm, LUDY-N and LUDY-S have shown excellent performance in various evaluation indexes, indicating that the proposed improvement strategies make the model have better robustness and generalization. Moreover, compared with multiple other popular competitors, the proposed improvement strategies enable LUD-YOLO to have the best overall performance, providing an effective solution for UAVs object detection while balancing model size and detection accuracy.

M‐SSD based on anchor proposal and ResNet101 backbone for placental haemorrhage MRI detection

AbstractMRI (magnetic resonance imaging) images can effectively show the placental haemorrhage area. In view of the special properties and real‐time detection requirements of placental haemorrhage MRI images, this paper has systematically improved the single‐shot multi‐box detector (SSD) target detection algorithm (M‐SSD). First, taking advantage of the particularity of the MRI image, the maximum stable extremum region (MSER) algorithm was used as the anchor proposal network which integrated the proposal information into the feature layer of SSD to avoid the hungry traversal of the original algorithm. Second, after the scale statistics of the placental haemorrhage area in MRI images, the bounding box matching the size of the placental haemorrhage area was redefined, in this way, the scale of the bounding box will have application pertinence, which can effectively improve the detection accuracy of the algorithm. Third, due to the small target property of the placental haemorrhage area in the MRI image, the VGG16 basic network in the original SSD was replaced by ResNet101, this made the algorithm have higher performance in small target detection. Finally, the Placental Haemorrhage MRI Detection Database (PHMD) has been built which is not only a base for this paper, but also for further research in this area.