- 标题
- 摘要
- 关键词
- 实验方案
- 产品
过滤筛选
- 2019
- 2018
- pattern recognition
- image
- partial discharge
- convolutional neural network(CNN)
- Conditional Random Fields (CRF)
- Convolutional Neural Network (CNN)
- Fine Classification
- Airborne hyperspectral
- green tide
- Elegant End-to-End Fully Convolutional Network (E3FCN)
- Optoelectronic Information Science and Engineering
- Shanghai Jiao Tong University
- Ocean University of China
- University of Oulu
- Wuhan University
- Central South University
- Hubei University
-
[IEEE 2019 IEEE Intelligent Vehicles Symposium (IV) - Paris, France (2019.6.9-2019.6.12)] 2019 IEEE Intelligent Vehicles Symposium (IV) - Circular Convolutional Neural Networks for Panoramic Images and Laser Data
摘要： Circular Convolutional Neural Networks (CCNN) are an easy to use alternative to CNNs for input data with wrap-around structure like 360? images and multi-layer laser scans. Although circular convolutions have been used in neural networks before, a detailed description and analysis is still missing. This paper closes this gap by de?ning circular convolutional and circular transposed convolutional layers as the replacement of their linear counterparts, and by identifying pros and cons of applying CCNNs. We experimentally evaluate their properties using a circular MNIST classi?cation and a Velodyne laser scanner segmentation dataset. For the latter, we replace the convolutional layers in two state-of-the-art networks with the proposed circular convolutional layers. Compared to the standard CNNs, the resulting CCNNs show improved recognition rates in image border areas. This is essential to prevent blind spots in the environmental perception. Further, we present and evaluate how weight transfer can be used to obtain a CCNN from an available, readily trained CNN. Compared to alternative approaches (e.g. input padding), our experiments show bene?ts of CCNNs and transfered CCNs regarding simplicity of usage (once the layer implementations are available), performance and runtime for training and inference. Implementations for Keras with Tensor?ow are provided online.
关键词： weight transfer,Circular Convolutional Neural Networks,CCNN,panoramic images,laser data
更新于2025-09-16 10:30:52
-
Robust Classification of Tea Based on Multi-Channel LED-Induced Fluorescence and a Convolutional Neural Network
摘要： A multi-channel light emitting diode (LED)-induced fluorescence system combined with a convolutional neural network (CNN) analytical method was proposed to classify the varieties of tea leaves. The fluorescence system was developed employing seven LEDs with spectra ranging from ultra-violet (UV) to blue as excitation light sources. The LEDs were lit up sequentially to induce a respective fluorescence spectrum, and their ability to excite fluorescence from components in tea leaves were investigated. All the spectral data were merged together to form a two-dimensional matrix and processed by a CNN model, which is famous for its strong ability in pattern recognition. Principal component analysis combined with k-nearest-neighbor classification was also employed as a baseline for comparison. Six grades of green tea, two types of black tea and one kind of white tea were verified. The result proved a significant improvement in accuracy and showed that the proposed system and methodology provides a fast, compact and robust approach for tea classification.
关键词： tea,variety,classification,convolutional neural network,EEM,LED,fluorescence
更新于2025-09-16 10:30:52
-
Dilated residual learning with skip connections for real-time denoising of laser speckle imaging of blood flow in a log-transformed domain
摘要： Laser speckle contrast imaging (LSCI) is a wide-field and noncontact imaging technology for mapping blood flow. Although the denoising method based on block-matching and filtering three-dimensional (BM3D) was proposed to improve its signal-to-noise ratio (SNR) significantly, the processing time makes it difficult to realize real-time denoising. Furthermore, it is still difficult to obtain an acceptable level of SNR with a few raw speckle images given the presence of significant noise and artifacts. A feed-forward denoising convolutional neural network (DnCNN) achieves state-of-the-art performance in denoising nature images and is efficiently accelerated by GPU. However, it performs poorly in learning with original speckle contrast images of LSCI owing to the inhomogeneous noise distribution. Therefore, we propose training DnCNN for LSCI in a log-transformed domain to improve training accuracy and it achieves an improvement of 5.13 dB in the peak signal-to-noise ratio (PSNR). To decrease the inference time and improve denoising performance, we further propose a dilated deep residual learning network with skip connections (DRSNet). The image-quality evaluations of DRSNet with five raw speckle images outperform that of spatially average denoising with 20 raw speckle images. DRSNet takes 35 ms (i.e., 28 frames per second) for denoising a blood flow image with 486 × 648 pixels on an NVIDIA 1070 GPU, which is approximately 2.5 times faster than DnCNN. In the test sets, DRSNet also improves 0.15 dB in the PSNR than that of DnCNN. The proposed network shows good potential in real-time monitoring of blood flow for biomedical applications.
关键词： Blood flow,convolutional neural network (CNN),laser speckle contrast imaging (LSCI),dilated convolution,skip connection
更新于2025-09-16 10:30:52
-
[IEEE IGARSS 2019 - 2019 IEEE International Geoscience and Remote Sensing Symposium - Yokohama, Japan (2019.7.28-2019.8.2)] IGARSS 2019 - 2019 IEEE International Geoscience and Remote Sensing Symposium - Extraction of a Specific Land-Cover Class from Very High Spatial Resolution Imagery Using Positive and Unlabeled Learning with Convolutional Neural Networks
摘要： In remote sensing, supervised multiclass classifiers show a very promising performance in terms of classification accuracy. However, they require that all classes, in the study area, are labeled. In many applications, users may only be interested in specific land classes. When considering only one class, this referred to as One-Class classification (OC) problem. In this paper, we investigated the possibility of using Convolutional Neural Networks (CNN) within the Positive and Unlabeled Learning (PUL) framework for estimating the urban tree canopy coverage from very high spatial resolution aerial imagery. We also compared the proposed approach to the Binary CNN classification and to ensemble classifications based on various color-texture based features. The obtained classification accuracies show that PUL strategies provide competitive extraction results, especially the proposed CNN based one, due to the fact that PUL is a positive-unlabeled method in which large amounts of available unlabeled samples is incorporated into the training phase, allowing the classifier to model effectively the tree class.
关键词： convolutional neural networks,texture analysis,One-class classification,positive and unlabeled learning,ensemble classification
更新于2025-09-16 10:30:52
-
Identification of Soybean Varieties Using Hyperspectral Imaging Coupled with Convolutional Neural Network
摘要： Soybean variety is connected to stress resistance ability, as well as nutritional and commercial value. Near-infrared hyperspectral imaging was applied to classify three varieties of soybeans (Zhonghuang37, Zhonghuang41, and Zhonghuang55). Pixel-wise spectra were extracted and preprocessed, and average spectra were also obtained. Convolutional neural networks (CNN) using the average spectra and pixel-wise spectra of different numbers of soybeans were built. Pixel-wise CNN models obtained good performance predicting pixel-wise spectra and average spectra. With the increase of soybean numbers, performances were improved, with the classification accuracy of each variety over 90%. Traditionally, the number of samples used for modeling is large. It is time-consuming and requires labor to obtain hyperspectral data from large batches of samples. To explore the possibility of achieving decent identification results with few samples, a majority vote was also applied to the pixel-wise CNN models to identify a single soybean variety. Prediction maps were obtained to present the classification results intuitively. Models using pixel-wise spectra of 60 soybeans showed equivalent performance to those using the average spectra of 810 soybeans, illustrating the possibility of discriminating soybean varieties using few samples by acquiring pixel-wise spectra.
关键词： a majority vote,convolutional neural network,hyperspectral imaging technology,soybean,pixel-wise spectra
更新于2025-09-16 10:30:52
-
[IEEE 2018 11th UK-Europe-China Workshop on Millimeter Waves and Terahertz Technologies (UCMMT) - HangZhou, China (2018.9.5-2018.9.7)] 2018 11th UK-Europe-China Workshop on Millimeter Waves and Terahertz Technologies (UCMMT) - RCS Reduction of 2??2 Microstrip Antenna Array Using All Dielectric Metasurface
摘要： In this paper, we investigate the feasibility of recognizing human hand gestures using micro-Doppler signatures measured by Doppler radar with a deep convolutional neural network (DCNN). Hand gesture recognition using radar can be applied to control electronic appliances. Compared with an optical recognition system, radar can work regardless of light conditions and it can be embedded in a case. We classify ten different hand gestures, with only micro-Doppler signatures on spectrograms without range information. The ten gestures, which included swiping from left to right, swiping from right to left, rotating clockwise, rotating counterclockwise, pushing, double pushing, holding, and double holding, were measured using Doppler radar and their spectrograms investigated. A DCNN was employed to classify the spectrograms, with 90% of the data utilized for training and the remaining 10% for validation. After ?ve-fold validation, the classi?cation accuracy of the proposed method was found to be 85.6%. With seven gestures, the accuracy increased to 93.1%.
关键词： Doppler radar,micro-Doppler signatures,Hand gesture,deep convolutional neural networks
更新于2025-09-16 10:30:52
-
Automatic seam detection and tracking system for robots based on laser vision
摘要： Detecting weld seams and accurately locating them from time-series images that contain strong noise pollution are di?cult and lead to reduced tracking accuracy. To address this problem, this paper presents a seam detection and tracking algorithm from the perspective of seam feature extraction and seam detection and positioning. The proposed image-processing algorithm employs the powerful feature expression capability and self-learning function of the deep convolutional neural network. A real-time weld seam searching and positioning strategy based on the multi-correlation ?lter cooperative detection mechanism is proposed in consideration of the continuity of the motion of the feature points of adjacent frames and the correlation of laser stripe structural information. Experimental results show that the sensor’s measurement frequency can reach 20 Hz, the average absolute tracking error of straight or curved welds is less than 0.25 mm, and the maximum tracking error does not exceed 1 mm in an environment with strong arc and splash noise. Moreover, the welding torch end runs smoothly during welding. The proposed strategy can meet high-quality welding requirements.
关键词： Correlation ?lter,Deep convolutional neural network,Laser vision sensor,Seam tracking
更新于2025-09-16 10:30:52
-
A convolutional approach to quality monitoring for laser manufacturing
摘要： The extraction of meaningful features from the monitoring of laser processes is the foundation of new non-destructive quality inspection methods for the manufactured pieces, which has been and remains a growing interest in industry. We present ConvLBM, a novel approach to monitor Laser Based Manufacturing processes in real-time. ConvLBM uses a Convolutional Neural Network model to extract features and quality indicators from raw Medium Wavelength Infrared coaxial images. We demonstrate the ability of ConvLBM to represent process dynamics, and predict quality indicators in two scenarios: dilution estimation in Laser Metal Deposition, and location of defects in laser welding processes. Obtained results represent a breakthrough in the 3D printing of large metal parts, and in the quality control of welding processes. We are also releasing the ?rst large dataset of annotated images of laser manufacturing.
关键词： Quality-control,Laser-cladding,Laser-welding,Convolutional-neural-networks,Neural-networks
更新于2025-09-16 10:30:52
-
Research of Multimodal Medical Image Fusion Based on Parameter-Adaptive Pulse-Coupled Neural Network and Convolutional Sparse Representation
摘要： Visual effects of medical image have a great impact on clinical assistant diagnosis. At present, medical image fusion has become a powerful means of clinical application. The traditional medical image fusion methods have the problem of poor fusion results due to the loss of detailed feature information during fusion. To deal with it, this paper proposes a new multimodal medical image fusion method based on the imaging characteristics of medical images. In the proposed method, the non-subsampled shearlet transform (NSST) decomposition is first performed on the source images to obtain high-frequency and low-frequency coefficients. The high-frequency coefficients are fused by a parameter-adaptive pulse-coupled neural network (PAPCNN) model. The method is based on parameter adaptive and optimized connection strength β adopted to promote the performance. The low-frequency coefficients are merged by the convolutional sparse representation (CSR) model. The experimental results show that the proposed method solves the problems of difficult parameter setting and poor detail preservation of sparse representation during image fusion in traditional PCNN algorithms, and it has significant advantages in visual effect and objective indices compared with the existing mainstream fusion algorithms.
关键词： multimodal,medical image fusion,parameter-adaptive pulse-coupled neural network,convolutional sparse representation,non-subsampled shearlet transform
更新于2025-09-16 10:30:52
-
Automatic road-marking detection and measurement from laser-scanning 3D profile data
摘要： Automatic road-marking detection and measurement have great significance for pavement maintenance and management. Laser-scanning 3D profile data provide a new way of road-marking detection and measurement with an elevation accuracy of about 0.25 mm. This paper presents an automatic road-marking detection and measurement method that uses laser scanning of 3D pavement data. The elevation characteristics and geometric statistics that characterize road markings have been fully analyzed using 3D data. The first step was to use a specially designed step-shaped operator to convolve profile data to identify the regions of suspected marking edges at the profile level, which helps reduce the influence of other pavement factors, including crosswise-slope information, cracks, and rutting. Next, by combining the geometric characteristics of the road-marking region and the continuity of the convolution features at image level, the regions of suspected 3D road markings were extracted. Third, a convolutional neural network was introduced to distinguish real-marking data more clearly. Finally, the three-dimension measurement information was extracted from the detected region and from elevation information. Road-marking recognition experiments were then conducted based on real measured 3D data. The detection accuracies were all greater than 90.8% for 4178 test samples from five road sections with different kinds of road markings. Furthermore, the repeatability of multiple measurement results for road-marking elevations from two selected road sections was about 95%, and the correlation of the obtained road-marking elevations with manually measured elevations was about 85.36% for 200 measurement points.
关键词： Road-marking detection,Laser scanning,Convolution,Three-dimension measurement information,Convolutional neural network
更新于2025-09-16 10:30:52