Juxing Di

Juxing Di

Auto-generated profile, based on publicly available information

Juxing Di

College of Information Engineering, Hebei University of Architecture, Zhangjiakou, China

Research Focus & Metrics

Academic Career Length

4Years (2023-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
14 May 2025 in IEEE Access
Verified
Journal Article
Open access
GLGuisuo Liu
Guisuo Liu
JDJuxing Di
Juxing Di
QWQing Wang
Qing Wang
YZYan Zhao
Yan Zhao
YYYang Yang
Yang Yang
Accurate pest identification is crucial for ensuring both high quality and high yield in rice production. This paper proposes RicePest-YOLO, a practical and generalizable model designed for agricultural pest detection, based on structural optimization and lightweight strategies applied to the YOLOv8 architecture. The model is specifically tailored to address the challenges of rice pest images, such as small insect size, complex backgrounds, and high inter-class similarity. In particular, it integrates ODConv and the BiFPN structure into the backbone and neck of YOLOv8n, enhancing the model’s ability to extract features from rice pest images. Subsequently, the Shape-IoU loss function is employed to enhance the model’s sensitivity and precision in target recognition. Furthermore, to facilitate the deployment of the new algorithm on embedded terminal devices, we applied Layer Adaptive Magnitude Pruning and Knowledge Distillation techniques for a lightweight model. The effectiveness of the proposed model was evaluated on common rice pest datasets. Experimental results demonstrate that the proposed model outperforms YOLOv8n, achieving an mAP@0.5 of 94.3%, and an mAP@0.5:0.95 of 67.3%, with 2.7% improvement in mAP@0.5 and 1.2% improvement in mAP@0.5:0.95. Comparison experiments with state-of-the-art models further demonstrated the proposed approach’s superiority. The model’s generalization and robustness were tested on the Pests dataset and the Paddy Rice Pest dataset. Results indicated a 2.6% and 3.0% improvement in mAP@0.5, and a 2.8% and 4.0% increase in mAP@0.5:0.95, respectively. Additionally, Experimental results show that the lightweight version of RicePest-YOLO reduced parameters by 48.1% and GFLOPs by 50%, with only a marginal decrease of 0.4% in mAP@0.5. The results validate that the proposed model is not only effective and efficient but also robust and lightweight, making it a promising solution for rice pest detection and offering valuable support for pest management in precision agriculture.

16 December 2024 in IEEE Access
Verified
Journal Article
Open access
YYYang Yang
Yang Yang
FFFangtao Feng
Fangtao Feng
GLGuisuo Liu
Guisuo Liu
JDJuxing Di
Juxing Di
In recent years, deep learning has been extensively applied to small object detection, achieving significant advancements. Nevertheless, there remains substantial potential to improve both the effectiveness and efficiency of small object detection in Uncrewed aerial vehicle (UAV) images. In this article, we propose a novel YOLO model with multi-scale, effective, and lightweight methods for traffic small object detection, termed MEL-YOLO. Initially, the improved model reconstructed the multi-scale network structure, the high-resolution feature enhancement network increases size from 40x40 to 160x160. Secondly, the improved model incorporates a decoupled detection head, which separates the classification task from the localization task, thereby improving detection accuracy while increasing detection complexity. Furthermore, we explore the Soft-NMS algorithm to effectively mitigate small object occlusion and reduce missed detection. Finally, this study emphasizes the design of a lightweight model by introducing a model pruning strategy along with the Ghost module. We have selected four C3 modules, each with a substantial number of output channels, to integrate with the Ghost module. This method results in fewer parameters, lower GFLOPs, and higher FPS. Experiment on the VisDrone2019 public dataset, compared to the baseline YOLOv5s, our model demonstrates an improvement of 8.5% in mAP@0.5 and 6.2% increase in mAP@0.5:0.95. Furthermore, there is a 55.7% reduction in parameters, a decrease of 7.0% GFLOPs, and an increase of 11.3% in FPS. To demonstrates the versatility and robustness of MEL-YOLO model, we conducted validation experiments on the SODA10M and CCTSDB datasets, where the mAP@0.5 improved by 1.9% and 1.2%, the FPS increased by 18 frame/s and 6 frame/s, respectively. These results align with our expectations, indicating that our model is a more effective and lightweight algorithm for dense and small objects detection in aerial images.

2 September 2025 in IEEE Access
Verified
Journal Article
Open access
JDJuxing Di
Juxing Di
KXKaikai Xi
Kaikai Xi
HNHuichang Niu
Huichang Niu
XWXiawei Wu
Xiawei Wu
YYYang Yang
Yang Yang
The swift advancement of UAV technology is accelerating the intelligent transformation of object detection methods. UAV applications now span both civilian and military sectors, playing a crucial role in scenarios such as aerial surveying, disaster response, and security monitoring. However, UAV image detection faces numerous challenges: targets are affected by flight altitude and viewing angle, leading to significant scale variations; dynamic flight results in constantly changing spatial layouts and complex backgrounds—such as dense vegetation and building textures—often weaken or obscure object features, making detection more difficult. To tackle these issues, we present a novel YOLOv8-based model tailored for small object detection in UAV images. In response to the prevalence of small targets in UAV images, the model enhances the multi-scale network structure and introduces the Large Separable Kernel Attention-enhanced Spatial Pyramid Pooling Fast Layer (SPPF-LSKA), improving the model’s capability to extract features from and respond to small objects. Furthermore, the introduction of the Minimum Point Distance IoU (MPDIoU) and Soft Non-Maximum Suppression (Soft-NMS) enables effective detection of overlapping and occluded targets. Experimental results on the VisDrone2019 dataset show that, compared with YOLOv8, the improved model increases mAP@0.5 by 7.6%, reduces parameter count by 23%, and has a model size of just 5MB. Compared with other current algorithms, The enhanced model delivers the best detection performance, highlighting the effectiveness of the proposed approach.

27 April 2026 in Measurement Science and Technology
Verified
Journal Article
Open access
PDF file
JDJuxing di
Juxing di
QWQing Wang
Qing Wang
YYYang Yang
Yang Yang
Object detection in remote sensing plays an important role in measurement-oriented applications such as emergency monitoring, precision agriculture, and traffic observation. However, complex backgrounds, dense target distributions, and small object sizes remain challenging for reliable perception. In this work, we propose CMM-DETR, an efficient Transformer-based detector that improves the RT-DETR baseline through three components: CSGVMNet, MRFusion, and MSFPN. Specifically, we structurally improve the EfficientVIM block by leveraging a gated channel selection mechanism and cross-stage partial (CSP)-style connections, and further propose the CSGVM block to construct a new backbone network, CSGVMNet. To enhance feature aggregation, we develop MRFusion, which leverages the cross-scale fusion property of MFM and adaptively reweights multi-level features to enable dynamic cross-scale interactions. We further introduce MSFPN, an improved multi-scale feature pyramid that promotes effective exchange between high-level context and low-level details. In addition, a high-resolution detection head is employed to better preserve spatial resolution while maintaining an adequate receptive field, improving localization and recognition for small objects in crowded scenes. On VisDrone2019, CMM-DETR improves mAP50 and mAP50:95 by 6.4% and 5.4%, respectively, while reducing parameters by 18.6%. Experiments on UAVDT and RSOD further demonstrate the robustness and generalization of the proposed method. Overall, CMM-DETR provides an accuracy efficiency balanced solution for practical remote sensing measurement scenarios.

22 September 2023
Verified
Conference Paper
Open access
YYYang Yang
Yang Yang
JDJuxing Di
Juxing Di
WZWencheng Zhang
Wencheng Zhang
FYFumei Yu
Fumei Yu
With the popularization and massive use of Internet, various types of attacks and abnormal traffic in campus networks are emerging. In this paper, based on NetFlow technology and convolutional neural network to analyze abnormal traffic in campus network, we add point group convolution and channel randomization on the basis of traditional neural network, and solve the problems of group convolution by shuffling different channels, which greatly reduces the amount of model operations and speeds up the speed of model classification. The improved neural network model achieves 93.8%. in accuracy, which greatly improves the accuracy of abnormal traffic classification. The method can realize the real-time collection of campus network traffic and realize the real-time monitoring of campus network condition through traffic analysis and comparison, which can strongly guarantee the security and stability of university network.