A Feature Extraction Method for Person Re-identification Based on a Two-branch CNN

Abstract A two-branch convolutional neural network (CNN) architecture for feature extraction in person re-identification (re-ID) based on video surveillance is proposed. Highly discriminative person features are obtained by extracting both global and local features. Moreover, an adaptive triplet loss function based on the original triplet loss function is proposed and is used in the network training process, resulting in a significantly improved learning efficiency. The experimental results on open datasets demonstrate the effectiveness of the proposed method.

Download Full-text

Aircraft Type Recognition in Remote Sensing Images Based on Feature Learning with Conditional Generative Adversarial Networks

Remote Sensing ◽

10.3390/rs10071123 ◽

2018 ◽

Vol 10 (7) ◽

pp. 1123 ◽

Cited By ~ 8

Author(s):

Yuhang Zhang ◽

Hao Sun ◽

Jiawei Zuo ◽

Hongqi Wang ◽

Guangluan Xu ◽

...

Keyword(s):

Remote Sensing ◽

Feature Extraction ◽

Loss Function ◽

Extraction Method ◽

Generative Adversarial Networks ◽

Support Vector ◽

Svm Classifier ◽

Feature Extraction Method ◽

Adversarial Networks ◽

Aircraft Type

Aircraft type recognition plays an important role in remote sensing image interpretation. Traditional methods suffer from bad generalization performance, while deep learning methods require large amounts of data with type labels, which are quite expensive and time-consuming to obtain. To overcome the aforementioned problems, in this paper, we propose an aircraft type recognition framework based on conditional generative adversarial networks (GANs). First, we design a new method to precisely detect aircrafts’ keypoints, which are used to generate aircraft masks and locate the positions of the aircrafts. Second, a conditional GAN with a region of interest (ROI)-weighted loss function is trained on unlabeled aircraft images and their corresponding masks. Third, an ROI feature extraction method is carefully designed to extract multi-scale features from the GAN in the regions of aircrafts. After that, a linear support vector machine (SVM) classifier is adopted to classify each sample using their features. Benefiting from the GAN, we can learn features which are strong enough to represent aircrafts based on a large unlabeled dataset. Additionally, the ROI-weighted loss function and the ROI feature extraction method make the features more related to the aircrafts rather than the background, which improves the quality of features and increases the recognition accuracy significantly. Thorough experiments were conducted on a challenging dataset, and the results prove the effectiveness of the proposed aircraft type recognition framework.

Download Full-text

ON-BOARD GCPS MATCHING WITH IMPROVED TRIPLET LOSS FUNCTION

ISPRS Annals of Photogrammetry Remote Sensing and Spatial Information Sciences ◽

10.5194/isprs-annals-v-2-2020-105-2020 ◽

2020 ◽

Vol V-2-2020 ◽

pp. 105-112

Author(s):

G. Xie ◽

Z. Zhang ◽

Y. Zhu ◽

S. Xiang ◽

M. Wang

Keyword(s):

Remote Sensing ◽

Feature Extraction ◽

Power Consumption ◽

Loss Function ◽

Intelligent Processing ◽

Single Frame ◽

Triplet Loss ◽

Consumption Model ◽

Remote Sensing Satellite ◽

Extraction Model

Abstract. Intelligent remote sensing satellite system is an important direction to solve the problem of intelligent processing on-board. It can realize the real-time on-board intelligent processing of important targets. The accuracy of geometric positioning information is the basis for subsequent intelligent processing. Therefore, this paper corrects the positioning information by GCPs (Ground Control Points) matching on-board. Considering the limited storage and computing performance of satellites, this paper designs a lightweight GCPs deep feature extraction convolutional neural network based on MobileNetV2 as feature extraction model, and trains this network with an improved triplet loss function. The Songshan calibration field images constructed by Wuhan University was used as the GCPs image, and 30,399 image patches were extracted and embedded as GCPs feature library. The size of the GCPs library is a size of 15.3M, and size of the lightweight depth feature extraction model is 9.83M, which can be pre-stored on the satellite for positioning with GCPs matching on-board. In addition, this paper tested feature extraction performance on an embedded device Nvidia Jeston Xavier which simulates the performance of the device on the satellite. At Xavier 30W max power consumption model, a single frame takes 0.005 seconds, and under Xavier 15W power consumption model, a single frame takes 0.009 seconds. At 10W power consumption model, a single frame takes 0.018 seconds, which can meet the performance requirements on the satellite. In addition, the experiments in this paper show that the positioning accuracy is within 30 meters. The work done in this paper will be experimented on the Luojia-3-01 intelligent remote sensing satellite.

Download Full-text

Instance Hard Triplet Loss for In-video Person Re-identification

Applied Sciences ◽

10.3390/app10062198 ◽

2020 ◽

Vol 10 (6) ◽

pp. 2198

Author(s):

Xing Fan ◽

Wei Jiang ◽

Hao Luo ◽

Weijie Mao ◽

Hongyan Yu

Keyword(s):

Loss Function ◽

Large Scale ◽

Video Camera ◽

Human Tracking ◽

Training Time ◽

Pose Tracking ◽

Person Features ◽

Potential Applications ◽

Triplet Loss ◽

Video Group

Traditional Person Re-identification (ReID) methods mainly focus on cross-camera scenarios, while identifying a person in the same video/camera from adjacent subsequent frames is also an important question, for example, in human tracking and pose tracking. We try to address this unexplored in-video ReID problem with a new large-scale video-based ReID dataset called PoseTrack-ReID with full images available and a new network structure called ReID-Head, which can extract multi-person features efficiently in real time and can be integrated with both one-stage and two-stage human or pose detectors. A new loss function is also required to solve this new in-video problem. Hence, a triplet-based loss function with an online hard example mining designed to distinguish persons in the same video/group is proposed, called instance hard triplet loss, which can be applied in both cross-camera ReID and in-video ReID. Compared with the widely-used batch hard triplet loss, our proposed loss achieves competitive performance and saves more than 30% of the training time. We also propose an automatic reciprocal identity association method, so we can train our model in an unsupervised way, which further extends the potential applications of in-video ReID. The PoseTrack-ReID dataset and code will be publicly released.

Download Full-text

FWNNet: Presentation of a New Classifier of Brain Tumor Diagnosis Based on Fuzzy Logic and the Wavelet-Based Neural Network Using Machine-Learning Methods

Computational Intelligence and Neuroscience ◽

10.1155/2021/8542637 ◽

2021 ◽

Vol 2021 ◽

pp. 1-13

Author(s):

Mohsen Ahmadi ◽

Fatemeh Dashti Ahangar ◽

Nikoo Astaraki ◽

Mohammad Abbasi ◽

Behzad Babaei

Keyword(s):

Neural Network ◽

Fuzzy Logic ◽

Feature Extraction ◽

Brain Tumor ◽

Tumor Diagnosis ◽

Training Process ◽

Feature Extraction Method ◽

Gradient Descent Algorithm ◽

Brain Tumor Diagnosis ◽

The Brain

In this paper, we present a novel classifier based on fuzzy logic and wavelet transformation in the form of a neural network. This classifier includes a layer to predict the numerical feature corresponded to labels or classes. The presented classifier is implemented in brain tumor diagnosis. For feature extraction, a fractal model with four Gaussian functions is used. The classification is performed on 2000 MRI images. Regarding the results, the accuracy of the DT, KNN, LDA, NB, MLP, and SVM is 93.5%, 87.6%, 61.5%, 57.5%, 68.5%, and 43.6%, respectively. Based on the results, the presented FWNNet illustrates the highest accuracy of 100% with the fractal feature extraction method and brain tumor diagnosis based on MRI images. Based on the results, the best classifier for diagnosis of the brain tumor is FWNNet architecture. However, the second and third high-performance classifiers are the DT and KNN, respectively. Moreover, the presented FWNNet method is implemented for the segmentation of brain tumors. In this paper, we present a novel supervised segmentation method based on the FWNNet layer. In the training process, input images with a sweeping filter should be reshaped to vectors that correspond to reshaped ground truth images. In the training process, we performed a PSO algorithm to optimize the gradient descent algorithm. For this purpose, 80 MRI images are used to segment the brain tumor. Based on the results of the ROC curve, it can be estimated that the presented layer can segment the brain tumor with a high true-positive rate.

Download Full-text

Evaluation of Fine Tuning and Feature Extraction methods in Biometric Periocular Recognition

10.5753/wvc.2019.7626 ◽

2019 ◽

Author(s):

William Barcellos ◽

Nicolas Hiroaki Shitara ◽

Carolina Toledo Ferraz ◽

Raissa Tavares Vieira Queiroga ◽

Jose Hiroki Saito ◽

...

Keyword(s):

Feature Extraction ◽

Transfer Learning ◽

Class Imbalance ◽

Extraction Methods ◽

Fine Tuning ◽

Feature Extraction Method ◽

Network Training ◽

Learning Techniques ◽

Lower Accuracy ◽

Different Characteristics

The aim of this paper is to evaluate the performance of Transfer Learning techniques applied in Convolucional Neural Networks for biometric periocular classification. Two aspects of Transfer Learning were evaluated: the technique known as Fine Tuning and the technique known as Feature Extraction. Two CNN architectures were evaluated, the AlexNet and the VGG-16, and two image databases were used. These two databases have different characteristics regarding the method of acquisition, the amount of classes, the class balancing, and the number of elements in each class. Three experiments were conducted to evaluate the performance of the CNNs. In the first experiment we measured the Feature Extraction accuracy, and in the second one we evaluated the Fine Tuning performance. In the third experiment, we used the AlexNet for Fine Tuning in one database, and then, the FC7 layer of this trained CNN was used for Feature Extraction in the other database. We concluded that the data quality (the presence or not of class samples in the training set), class imbalance (different number of elements in each class) and the selection method of the training and testing, directly influence the CNN accuracy. The Feature Extraction method, by being more simple and does not require network training, has lower accuracy than Fine Tuning. Furthermore, Fine Tuning a CNN with periocular's images from one database, doesn't increase the accuracy of this CNN in Feature Extraction mode for another periocular's database. The accuracy is quite similar to that obtained by the original pre-trained network

Download Full-text