ZZZhifeng Zhou

Zhifeng Zhou

*This is an algorithmically generated author record based on publicly available information.

Zhifeng Zhou has a total of 34 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • 基于局部特征聚合与驱动点云采样的点云分类与分割

    by Huang Shuqi, Zhou Zhifeng, Ren Pulin, Zhai Chuanrun

    针对目前大多数深度学习模型获取局部特征的能力比较欠缺的问题,提出了一种基于局部特征聚合与驱动点云采样的点云分类与分割网络——FCA-PointNet++。设计了局部特征聚合模块,在该模块中构建了增强相关局部特征并抑制不相关特征的FCA(feature channel attention)模块;设计了一种根据不同任务改变初始采样点分布的驱动点云采样网络(drive point cloud sampling,DPN);构造了一种能准确提取局部特征并且自适应细化特征的通道门(channel-wise gate)模块。在分类任务中,FCA-PointNet++在ModelNet10和ModelNet40上的总体准确率(overall accuracy,OA)分别达到了95.3%、93.4%。在分割任务中,FCA-PointNet++的平均交并比(mean intersection over union,MIoU)达到了86.1%,实验验证了FCA-PointNet++在分类和分割任务中的有效性。

  • 基于八叉树与3D-SIFT特征提取的点云配准算法

    by Shuqi Huang, 上海工程技术大学机械与汽车工程学院,上海201620, Zhifeng Zhou, Pulin Ren, Liduan Wang, 空装驻常州地区军事代表室,江苏常州213100, 上海司南卫星导航技术股份有限公司产业研究院,上海201800

  • Open Access

    YOLOv11-BSS: Damaged Region Recognition Based on Spatial and Channel Synergistic Attention and Bi-Deformable Convolution in Sanding Scenarios

    by Yinjiang Li, Zhifeng Zhou, Ying Pan

    In order to address the problem that the paint surface of the damaged region of the body is similar to the color texture characteristics of the usual paint surface, which leads to the phenomenon of leakage or misdetection in the detection process, an algorithm for detecting the damaged region of the body based on the improved YOLOv11 is proposed. Firstly, bi-deformable convolution is proposed to optimize the convolution kernel shape offset direction, which effectively improves the feature representation power of the backbone network; secondly, the C2PSA-SCSA module is designed to construct the coupling between spatial attention and channel attention, which enhances the perceptual power of the backbone network, and makes the model pay better attention to the damaged region features. Then, based on the GSConv module and the DWConv module, we build the slim-neck feature fusion network based on the GSConv module and DWConv module, which effectively fuses local features and global features to improve the saturation of semantic features; finally, the Focaler-CIoU border loss function is designed, which makes use of the principle of Focaler-IoU segmented linear mapping, adjusts the border loss function’s attention to different samples, and improves the model’s convergence of feature learning at various scales. The experimental results show that the enhanced YOLOv11-BSS network improves the precision rate by 7.9%, the recall rate by 1.4%, and the mAP@50 by 3.7% over the baseline network, which effectively reduces the leakage and misdetection of the damaged areas of the car body.


  • Open Access

    3D point cloud segmentation algorithm based on fused DenseNet and PointNet

    by Wu Liequan, Zhou Zhifeng, Shi Yun, Ren Pulin, 上海工程技术大学机械与汽车工程学院,上海201620, 上海航天设备制造总厂有限公司,上海 200245, 空装驻常州地区军事代表室,江苏常州213100, Ltd. Shanghai Aerospace Equipment Manufacture Co.

    点云分割对于智能驾驶、物体检测和识别、逆向工程等任务非常重要。PointNet是一种能够直接处理点云数据的方法,近年来在点云分割任务中得到广泛应用,但其分割精度较低,而PointNet++的计算成本又较高。针对以上问题,提出一种融合DenseNet和PointNet的算法,用于点云分割,并引入三分支混合注意力机制,以提高PointNet在提取局部特征方面的能力。基于密集连接卷积网络(DenseNet)思想,提出用DenseNet-STN和DenseNet-MLP结构来替代PointNet中的空间变换网络(STN)和多层感知机(MLP);同时,使用Add连接代替密集块(DenseBlock)中的Concat连接,以提高对点特征间相关性的准确性,同时不显著增加模型复杂度。DenseNet-PointNet能够提高复杂分类问题的泛化能力,实现对复杂函数更好的逼近,从而提高点云分割的准确率。有效性和消融实验结果表明,本文算法具有良好的性能。点云分割实验结果表明,DenseNet-PointNet在大多数类别中的交并比(IoU)都高于PointNet的IoU,并在部分类别中也高于PointNet++,参数量是PointNet++的47.6%,浮点运算量(FLOPs)是PointNet++的49.1%。实验结果验证了DenseNet-PointNet的可行性和有效性。


  • Open Access

    Automatic Measurement of Scoliosis Based on an Improved Segmentation Model

    by Jia Zhu, Zhifeng Zhou, Chengxian Yao

    The Cobb angle is an important indicator for evaluating the severity of spinal curvature. Currently, there are various segmentation models used to segment the spine on X-ray images, but there is still room for improvement in segmentation performance. In addition, existing automatic methods for detecting the Cobb angle require manual cropping or bounding of the dataset, and cannot achieve fully automatic detection. Therefore, this paper proposes an automatic detection method based on RetinaNet and W-Net, which detects the ROI of the spine and performs segmentation to achieve automatic detection of the 17 required vertebrae. Experimental results show that the MIoU of W-Net is 0.9084±0.0032, the Dice Coefficient is 0.9442±0.0161, and the Precision is 0.9433±0.0176, which is better than other typical segmentation models. In addition, a feasible post-processing method is proposed to solve the problem of adjacent vertebrae adhesion in the segmentation image. The reliability of the Cobb angle automatic detection method was evaluated, and the intra-group correlation coefficients (ICC) with two observers were 0.902 and 0.915, respectively, both greater than 0.9, and the mean absolute deviation (MAD) was 3.08° and 2.91°, respectively, both less than 3.5°. Therefore, the proposed method can achieve automatic detection of the Cobb angle without manual cropping or additional human intervention, and has good reliability.


  • Open Access

    Vertebral Center Points Locating and Cobb Angle Measurement Based on Deep Learning

    by Zhifeng Zhou, Jia Zhu, Chengxian Yao

    Traditional manual measurement of Cobb angle is a time-consuming process and leads to different results. To address this issue, this paper proposes a deep learning-based method of locating the vertebral center points. The whole X-ray can be input into the network for prediction, without worrying about the detection of cervical vertebrae with similar characteristics to the thoracic and lumbar vertebrae. First, key points predicting and noise points filtering operations are employed to obtain vertebral center points for fitting. Then, the spine curve is fitted, and the slope of the normal line of the spine curve is adjusted according to an empirical formula. Finally, the Cobb angle allowed by the error is calculated. Through the reliability analysis of the traditional manual measurement method and the automatic detection method in this paper, ICC (intraclass correlation coefficient) with the two observers was 0.897 and 0.901, respectively, and MAD (mean absolute deviation) was 3.13° and 3.04° respectively. This indicates that the automatic detecting method by computer has good reliability. Therefore, this method can be used to detect scoliosis quickly and effectively.


  • Surface defect detection of patch diode based on improved YOLO-V4

    by Wu Liequan, Zhou Zhifeng, Zhu Zhiling, Zhang Wei, Wang Yong, 上海工程技术大学机械与汽车工程学院,上海201620, 上海司南卫星导航技术股份有限公司,上海 201801, 国网思极位置服务有限公司,北京 102211, Ltd. State Grid Siji Location Service Co.

    Aiming at low efficiency of traditional visual detection method and shallow model as well as low semantic character of target detection algorithm based on manual features, an surface defect detection method of patch diode based on improved YOLO-V4 was proposed. Firstly, DenseNet was used in CSP1 module to replace ResNet in original network, considering that gradient disappeared with network deepening and feature redundancy as well as parameters were reduced. Then, to realize cross-dimensional interaction of feature information and make the network pay more attention to important information, the three-branch attention mechanism module was introduced after CSP1 module, and features were fused with FPN+PANet. CBL×5 module was replaced by CSP2, which reduced computation of network and improved detection speed of algorithm. Finally, the Focal Loss function was optimized and weight was added to positive and negative samples to solve the imbalance problems. The detection precision (P), recall ratio (R) and mean average precision (mAP) of the algorithm are 2.98%, 2.65% and 2.92% higher than that of YOLO-V4, respectively, which shows that the improved YOLO-V4 can effectively detect the surface defects of patch diode.


  • Open Access

    Visual Recognition and Localization of Printed Circuit Board

    by Yi Zhang, Zhifeng Zhou, Lei Zheng, Zina Zhu, Peng Xiao

    Precise positioning is one of the key steps in defect detection for printed circuit boards. This paper proposes a method to evaluate the roundness of printed circuit board image by least squares method and determine the position of the target circle by area sorting and screening method to realize the accurate extraction of the positioning parameters of the circular reference point (positioning circle). The CCD camera is used to capture images of the printed circuit board, and the median filtering is used for image preprocessing. The printed circuit board is separated from the image background by the gray projection segmentation technique, and the object in the image smaller than the positioning circle area is hidden. After comparing the circularity of the graphic in the printed circuit board image, two positioning circles are found according to the size of the graphic area to determine the coordinates of the center of the circle. The experimental results show that the positioning circle detection result of this method is accurate and requires less time than the traditional Hough transform detection circle.