DLDawei Li

Dawei Li

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

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Dawei Li has a total of 24 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Automated Scrap Steel Grading via a Hierarchical Learning-Based Framework

    by Qifan Tu, Dawei Li, Qian Xie, Li Dai, Jun Wang

    Scrap steel grading is relatively important during the process of scrap recycling. However, the current evaluation method for scrap steel grading mainly relies on the manual, which is subjective and inaccurate. There are two challenges for scrap steel grading, including detecting scrap steel precisely and evaluating its grade accurately. To address these issues, we propose a novel framework, including a carriage-attention module (CaM), a scrap detection module (SDM), and a scrap grading module (SGM). Owing to its complex background and large size range, it is a challenging task to detect scrap steel. To this end, the CaM is proposed to remove the complex background from the original scrap steel images. The SDM is then proposed to detect scrap steel in the carriage-only images that the complex background is removed. Specifically, the SDM is a segmentation network with a multi-scale feature fused pyramid (MFFP). Finally, based on a weight-balanced Bayes classifier, the SGM is developed to evaluate the final grades for all images captured from a carriage during the process of scrap unloading. Experimental results show that our approach obtains a high accuracy for scrap steel grading while ensuring a fast inference speed. Note that our method has been successfully applied in several steel smelters.


  • Yarn Density Measurement for 3-D Braided Composite Preforms Based on Rotation Object Detection

    by Li Dai, Dawei Li, Qifan Tu, Jun Wang

    This article examines how to accurately measure yarn density for 3-D braided composite preforms in large images. However, due to the complex structure and dense yarns of the preform, there is a lack of effective methods to solve this problem in practical applications. In the end, the task of yarn density measurement is mainly performed by experienced operators. They usually use a ruler to measure the distance between a fixed number of fabrics to calculate density, which is time-consuming and subjective. Therefore, an effective and efficient method for yarn density measurement of 3-D braided composite preforms is demanded. To address this problem, we propose a novel framework combining deep learning with traditional image processing techniques to measure yarn density automatically. First, to ensure the accuracy of yarn detection, we obtain a high-resolution raw image of $2448\times2048$ . The preprocessed images are then fed into our Yarn Detection Network (YDNet) based on the rotation object detection method to detect per segmented yarn. To achieve this, we propose a dilated feature network (DFN) that can obtain features at different scales and improve detection speed. Then, the feature alignment module (FAM) is introduced to achieve higher detection accuracy. Moreover, we design a Yarn Loss (YLoss) to detect yarn boxes more accurately via considering the aspect ratio of the yarns as an important parameter. Then, we design a multilevel rotation mechanism based on the extracted yarns to correct the image and calculate the yarn density accurately. We conduct several experiments to verify the superiority of our proposed YDNet, and the accuracy and effectiveness of the proposed framework for the automatic yarn density measurement. The experimental results show that our method is superior to other state-of-the-art methods.


  • Automatic Vision-Based Online Inspection System for Broken-Filament of Carbon Fiber With Multiscale Feature Learning

    by Dawei Li, Shiyan Hua, Zhongyu Li, Xiaoxi Gong, Jun Wang

    This article aims at solving the problem of broken-filament online detection and tracking during the process of carbon fiber production. Owing to the complex background and unique broken-filament shape, it is challenging to achieve the broken-filament online detection with high accuracy via classical machine vision methods, even general convolutional neural network (CNN)-based methods. Moreover, the carbon fiber strands are usually merged during the winding process, which would lead to the broken-filament location and tracking errors due to wrong segmentation and numbering of carbon fiber strands. To address these problems, we propose a vision-based method to achieve broken-filament online detection and tracking. First, an image collection module is designed to capture high resolution carbon fiber strand data. Then, a novel broken-filament detection network (BFDNet) is proposed to detect broken-filaments, including a feature learning part and a new anchor generation scheme, which is implemented based on the RetinaNet framework. Especially, the feature learning part contains several multiscale kernel fusion blocks (MKFBs) that are composed of convolution kernels of different sizes. Finally, a pixel projection method is proposed to segment and number carbon fiber strands for tracking. Our method achieves satisfactory performance for broken-filament online detection in terms of accuracy compared with several state-of-the-art methods designed for feature learning based on the classical object detection framework. Moreover, our method has been applied in practical applications for broken-filament inspection tasks in factories.


  • Accurate and Reliable Sealant Inspection for Aircraft Fuel Tank Based on 3-D Point Cloud

    by Xiaoxi Gong, Xuanming Cao, Dawei Li, Yabin Xu, Qian Xie, Qiaoyun Wu, Jun Wang

    As an important force-bearing component, aircraft fuel tank is localized at wings, fuselage, and empennage of an aircraft. To prevent the leakage or potential hazard of an aircraft fuel tank, the seal quality of fasteners on the fuel tank skin must be carefully inspected. Due to the limited space during inspection and its complex internal structures, it is challenging for current techniques to effectively inspect aircraft fuel tank. In this article, we introduce an aircraft sealant inspection system (ASInspS) to achieve accurate and reliable internal inspection for fuel tanks. For the hardware part, we utilize a high-precision 3-D scanner and a visual–inertial odometry (VIO) to acquire point clouds and poses, simultaneously. We then design three key modules to achieve the sealant inspection with high accuracy and reliability. First, the scanner positioning module aims to acquire the 6-D poses of the 3-D scanner, in which we propose two key improvements, that is, event-based mask (EvMask) and adaptive keyframe selection, to improve the robustness of feature tracking. Second, the seal area reconstruction module follows a coarse-to-fine strategy to achieve the accurate registration and then to create high-quality point cloud models of fuel tanks. Meanwhile, the results of fine alignment will be used to correct the accumulated drift error in the first module, which allows the system to ensure the high accuracy under complex scenes. Third, the sealant inspection module is built to conduct the accurate and reliable sealant measurement. The quantitative results demonstrate that our sealant inspection system can guarantee the precision (on different scales) at whatever the stage of the positioning, registration, or index measurement. The results also indicate that our ASInspS performs with high reliability even under a complex and closed aircraft fuel tank.


  • Integrated Circuit Gold Wire Bonding Measurement Via 3-D Point Cloud Deep Learning

    by Qian Xie, Kun Long, Dening Lu, Dawei Li, Yuan Zhang, Jun Wang

    In the semiconductor manufacturing industry, gold wires are commonly used to connect integrated circuits for transmitting signal and power. Hence, gold wire bonding quality inspection is a vital task for the functional stability of the integrated circuit chip. The shape of each gold wire bonding is one of the most important factors influencing the stability of chip performance. However, as the size of gold wire is quite small (around 0.02 mm), it is difficult to quantitatively measure the shape, i.e., the size-related indices. In this article, we propose a novel framework for fully automated gold wire bonding size-related measurement, which is the key step in gold wire bounding quality inspection. First, we employ a 3-D scanning device to obtain the 3-D point cloud of the chip area where gold wire bonding structures are. We then propose a 3-D deep learning-based algorithm to extract multiple gold wire bounding structures from the whole scanned point cloud. With the extracted points, a postprocessing method is introduced to group the gold wire-solder joint pairs. Moreover, we design a point cloud-based calculation model to perform the final size-related measurement. The proposed automated framework achieves high-accuracy quantitative measurement of gold wire bonding structures on the scale of 0.02 mm for the first time. The effectiveness of the proposed measurement framework is demonstrated by the experiments.


  • Tiny Defect Detection in High-Resolution Aero-Engine Blade Images via a Coarse-to-Fine Framework

    by Dawei Li, Yida Li, Qian Xie, Yuxiang Wu, Zhenghao Yu, Jun Wang

    This article studies the problem of aero-engine blade surface defect detection in large images. The effective method for aero-engine blade surface inspection for this real application is currently lacking since most defects are relatively small. Therefore, the task of aero-engine blade surface defect detection is mainly implemented by experienced operators, which is subjective and time-consuming. Moreover, it is hard to fit the requirements of higher accuracy and efficiency manually. Therefore, an effective and efficient method for aero-engine blade surface defect detection is demanded. To achieve this, we propose a vision-based framework in this study to detect defects in a coarse-to-fine manner. First, the captured raw images are with a high resolution of 2448×2048 to ensure the accuracy of defect detection. The raw images are then cropped into smaller regions and fed into our deep convolution neural network (DCNN) to learn features with high representation. Next, the coarse classifier module is proposed to filter most background regions out. Finally, the defects are located and classified by a fine detector module in the defective images, which are selected by the coarse classifier module. Instead of directly applying a detector, our coarse-to-fine framework can effectively save computation and improve accuracy. In addition, the coarse-to-fine framework can be trained in an end-to-end manner. Compared with classical methods for object detection, our method also achieves state-of-the-art performance for aero-engine blade surface defect detection in terms of accuracy and efficiency. Furthermore, our framework has been applied for practical application in many aero-engine blade production lines.


  • Detecting Trees in Street Images via Deep Learning With Attention Module

    by Qian Xie, Dawei Li, Zhenghao Yu, Jun Zhou, Jun Wang

    Although object detection techniques have been widely employed in various practical applications, automatic tree detection is still a difficult challenge, especially for street-view images. In this article, we propose a unified end-to-end trainable network for automatic street tree detection based on a state-of-the-art deep learning-based object detector. We tackle low illumination and heavy occlusion conditions in tree detection, which have not been extensively studied until now, due to clear challenges. Existing generic object detectors cannot be directly applied to this task due to aforementioned challenges. To address these issues, we first present a simple, yet effective image brightness adjustment method to handle low illuminance cases. Moreover, we propose a novel loss and a tree part-attention module to reduce false detections caused by heavy occlusion, inspired by the previously proposed occlusion-aware region-convolutional neural network (R-CNN) work. We train and evaluate several versions of the proposed network and validate the importance of each component. It is demonstrated that the resulting framework, part attention network for tree detection (PANTD), can efficiently detect trees in street-view images. The experimental results show that our approach achieves high accuracy and robustness under various conditions.