A target detection method for dense scene fusion multi-scale spatial pyramid

Abstract

The application provides an end-to-end trained target detection algorithm MSSP-YOLOv8. A multi-scale space pyramid (MSSP) is designed to capture the local features of the target and the context information of the surrounding environment; based on the idea of reparameterization, a C2fD module is designed by using a diverse branch block (DBB) to further improve the detection accuracy; the MSSP and the C2fD module are combined with YOLOv8 to construct the MSSP-YOLO target detection algorithm. The application can more effectively utilize the information of dense target image data and improve the detection accuracy of dense targets.

Description

Technical Field

[0001] This invention relates to the field of computer vision, specifically to a method for object detection in dense scenes that integrates multi-scale spatial pyramids. Background Technology

[0002] Object detection, a core foundational task in computer vision, plays a crucial role in image understanding. In recent years, deep learning technology has made significant progress, driving the rapid development of object detection methods and leading to a large number of deep learning-based approaches. These methods can be broadly categorized into two types: two-stage object detection algorithms and single-stage object detection algorithms. However, in complex real-world scenarios, especially those with densely distributed objects, occlusion and overlap between objects reduce the visible area and visual size of each object, increasing the difficulty of achieving high-precision detection.

[0003] Many scholars both domestically and internationally have conducted research on object detection, with deep learning and the introduction of attention mechanisms becoming important approaches. To address the problem of small, densely distributed targets in complex scenes, a multi-scale spatial pyramid structure can effectively capture contextual information at different levels, thereby significantly improving the accuracy of detection, segmentation, and classification, demonstrating outstanding performance in small object detection. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention provides a target detection method for dense scenes that integrates multi-scale spatial pyramids. By designing a multi-scale spatial pyramid and using the DBB module to construct a C2fD module, the detection performance is improved.

[0005] To achieve the aforementioned objectives, the present invention employs a method for target detection in dense scenes that integrates multi-scale spatial pyramids, comprising the following steps: Step S1: Design a multi-scale space pyramid (MSSP) to capture the local features of the target and the contextual information of the surrounding environment, thereby improving the identification and accurate detection of the target.

[0006] Step S2: Based on the concept of reparameterization, the C2f module is improved using multi-branch blocks (DBB) to construct the C2fD module. By integrating multi-scale and multi-complexity branching structures, the richness of the feature space is expanded, thereby improving the expressive power of a single convolutional layer and increasing the complexity and diversity of the model.

[0007] Step S3: The MSSP is embedded between the Backbone layer and Neck layer of YOLOv8, preserving the differences between objects while achieving information fusion; the C2fD module uses multi-parameter branching in the Backbone stage to improve target detection accuracy. Compared with other mainstream algorithms, the algorithm achieves higher detection accuracy.

[0008] The beneficial effects of the target detection method based on the fusion of multi-scale spatial pyramids for dense scenes proposed in this invention are as follows: This invention is based on YOLOv8, and designs a multi-scale spatial pyramid and a C2fD module. It combines the MSSP and C2fD modules with YOLOv8 to construct the MSSP-YOLO object detection algorithm. Experiments were conducted on the Kaggle_Mask dataset. The results show that MSSP-YOLO achieves higher detection accuracy compared with other mainstream algorithms, with an average accuracy of 90.4%. Attached Figure Description

[0009] Figure 1 shows the overall network structure of the target detection method for dense scenes using a fusion of multi-scale spatial pyramids according to the present invention. Figure 2 is a comparison of the detection performance of the target detection method of the fusion multi-scale spatial pyramid for dense scenes according to the present invention; Detailed Implementation Method 1 To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0010] As shown in Figure 1, a target detection method for dense scenes using a fusion of multi-scale spatial pyramids includes the following steps: Step S1: Select an object detection data instance: the Kaggle_Mask dataset.

[0011] Step S2: To construct the MSSP-YOLO detection algorithm, after integrating the pyramid structure of the MSSP module into the neck, the C2f module is improved using the multi-branch block DBB, and the C2fD module is proposed.

[0012] Step S3: Input the dataset from Step S1 into the MSSP-YOLO constructed in Step S2 and train it. The final average accuracy reached 90.4%. Detailed Implementation Method 2 To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0014] As shown in Figure 2, in order to verify the effectiveness of the core module, an ablation experiment was conducted, and the steps are as follows: Step S1: Based on the environment and Kaggle_Mask dataset of Implementation Method 1, construct 3 sets of models: full MSSP-YOLO, MSSP removed, and C2fD removed.

[0015] Step S2: Keep the training parameters consistent, train synchronously with the input dataset, and record the changes in accuracy.

[0016] Step S3: The results are shown in Table 1. The MSSP-YOLO model performed best, with a significant improvement in accuracy, confirming the effectiveness of the collaborative performance enhancement of each module.

[0017] To verify the effectiveness of the target detection method for dense scenes using a fusion of multi-scale spatial pyramids provided by this invention, the present invention also provides the following experiments.

[0018] (1) Experimental data and experimental environment The present invention uses the Kaggle_Mask dataset for training, with a training set to test set ratio of 8:2.

[0019] The experiment used Windows 11 operating system, NVIDIA RTX 3060 graphics card, Python 3.8, and PyTorch deep learning framework.

[0020] (2) Experimental setup and evaluation metrics The input size of all training images was uniformly adjusted to 640×640. The batch size was set to 32 and the number of training epochs was set to 100.

[0021] This invention selects mean average precision (mAP), number of parameters (params), computational cost (Giga Floating Point Operations, GFLOPs), and frames per second (FPS) as evaluation metrics.

[0022] The specific calculation formula is as follows.

[0023]

[0024] (3) Experimental results and analysis The performance comparison of the method of this invention with other models is shown in Table 1.

[0025]

[0026] Experimental results show that on the Kaggle_Mask mask-wearing dataset, MSSP-YOLO achieves improvements in both mAP50-95 and mAP compared to other state-of-the-art detection algorithms, indicating that MSSP-YOLO is suitable for target detection in dense scenes.

[0027] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A target detection method for dense scenes using a fusion of multi-scale spatial pyramids, characterized in that: S1: Construct a multi-scale space pyramid (MSSP) to capture the local features of the target and the contextual information of the surrounding environment. S2: Based on the idea of ​​reparameterization, the C2fD module was designed using the Diverse Branch Block (DBB). S3: The MSSP is embedded between the Backbone layer and Neck layer of YOLOv8 to achieve information fusion while preserving the differences between objects; the C2fD module uses multi-parameter branching in the Backbone stage to improve target detection accuracy.

2. The target detection method for dense scenes using a fusion of multi-scale spatial pyramids as described in claim 1, characterized in that, The multi-scale spatial pyramid described in step S1 is composed of context aggregation blocks with residual connections. Each layer of the multi-scale spatial pyramid is composed of a context information aggregation block with residual connections. The context information aggregation block is composed of convolutional layers for context refinement, attention map generation, and feature mapping.

3. The target detection method for dense scenes using a fusion of multi-scale spatial pyramids as described in claim 1, characterized in that, In step S2, the C2fD module inputs data through a convolutional layer during forward propagation. The output of the previous layer is divided into two paths. One path is directly passed to the output, and the other path is further processed by Bottleneck_DBB. Finally, the data from the two paths are merged along the channel dimension and integrated through another convolutional layer to form the final output of the network. The Bottleneck_DBB module operates in parallel with multiple branches during the training phase and is transformed into a single convolution during the inference phase using parameter fusion technology.

Images