Detection of dim and small infrared targets based on the most appropriate contrast saliency analysis
Jiang Guoqing, and Wan Lanjun
Aiming at the current situation of dim target detection in infrared image, a dim and small target detection method based on the most appropriate contrast saliency analysis was proposed. In sliding serial port, the non-linear processing technology was used to process the image, which avoided the saliency produced by traditional saliency analysis algorithm when processing the image at the scene edge. The problem of value interference does not affect the ability of target extraction in the target area. A large number of hardware-in-the-loop simulation experiments were carried out. The results show that although the proposed method can not improve significantly in the background suppression factor, the performance of target detection in the two indicators of mean signal-to-noise ratio and signal-to-noise ratio gain of the proposed method are significantly enhanced. And among the three methods of image processing, the effect is the best.
  • Jul. 30, 2021
  • Infrared and Laser Engineering
  • Vol.50 Issue, 4 20200377 (2021)
  • DOI:10.3788/IRLA20200377
Parameter estimation of attributed scattering centers in SAR images based on wolf pack algorithm
Chen Cong
For the attributed scattering center estimation problem of synthetic aperture radar (SAR) images, a new idea based on wolf pack algorithm was proposed. The method first decoupled the SAR image using the “divide and conquer” strategy. Afterwards, the wolf pack algorithm was adopted as the basic optimization method to sequentially estimate individual scattering centers with the optimal parameter sets. By analyzing the characteristics of cooperative hunting activities and prey distribution, the wolf pack algorithm has good global search ability and local development ability. The algorithm combines the traditional image decoupling with the robust estimation capability of wolf pack algorithm. Hence, the estimation precision of the overall SAR image can be improved. In the experiments, the proposed method was tested on the original SAR images and noisy samples based on the MSTAR dataset. The results validate the effectiveness and noise-robustness of the proposed method.
  • Jul. 30, 2021
  • Infrared and Laser Engineering
  • Vol.50 Issue, 4 20200409 (2021)
  • DOI:10.3788/IRLA20200409
Multivariate empirical mode decomposition with application to SAR image target recognition
Wu Youlong
A synthetic aperture radar (SAR) target recognition method was proposed based on multivariate empirical mode decomposition (MEMD). MEMD was the general extension of traditional EMD, which could avoid the mode mixing problems. MEMD was employed to process SAR images to obtain the multi-layer intrinsic mode functions (IMF), which could better reflect the time-frequency properties of the targets. Different layers of IMFs could effectively complement each other while sharing some inner correlations because they are generated from the same target. In the classification phase, the joint sparse representation was employed to represent the IMFs. The joint sparse representation could solve several related sparse representation tasks based on the idea of multi-task learning. It could produce more precise estimations than the solutions of single tasks. According to the sparse coefficient vectors corresponding to different IMFs, the reconstruction errors of different classes for the representation of the test sample can be calculated. Afterwards, the target label of the test sample can be determined. Experiments were conducted on the Moving and Stationary Target Acquisition and Recognition (MSTAR) dataset, by comparison with existing methods under the standard operating condition, depression angle variance, noise corruption, and target occlusion, the results confirm the validity of the proposed method.
  • Jul. 30, 2021
  • Infrared and Laser Engineering
  • Vol.50 Issue, 4 20200236 (2021)
  • DOI:10.3788/IRLA20200236
Air target reference spectrum selection based on characteristic wavelengths extracted by successive projections algorithm
Zhang Yunke, Ren Dengfeng, Han Yuge, and Li Jiyuan
Possessing certain spectral radiation characteristics in a relatively stable state, the air target can be identified according to its spectrum. At first, with the simulation model of air target spectral radiation, the spectral radiance was calculated. Secondly, successive projections algorithm was applied to extract the characteristic wavelengths from the simulation spectral data to reduce required data while retaining a certain accuracy. At last, the hybrid spectral similarity measure named SID (TAN) was involved in comparing the spectral radiation characteristics of different flying heights and flying time in 3-5 μm band and 8-14 μm band called dual atmospheric windows due to its stronger discrimination capability. The result shows that flying heights exert a greater effect on target spectral radiation characteristics than flying time. Meanwhile, changes in 3-5 μm band are more obvious than in 8-14 μm band. Therefore, aiming to improve the recognition accuracy, more factors are supposed to be considered in 3-5 μm band than in 8-14 μm. Compared with flying time, it is recommended to select more spectra of various flying heights as reference spectrum.
  • Jul. 30, 2021
  • Infrared and Laser Engineering
  • Vol.50 Issue, 4 20200250 (2021)
  • DOI:10.3788/IRLA20200250
Palmprint Recognition Based on Multi-Scale Gabor Orientation Weber Local Descriptors
Li Mengwen, Liu Huaiyu, Gao Xiangjun, and Meng Qianqian
Weber local descriptor (WLD) is an effective image feature descriptor. However, the differential excitation and gradient orientation, which are two components of WLD, can not accurately describe the difference of local image blocks and the orientation of palm lines, so the performance of palmprint recognition based on WLD features is not high. In order to improve palmprint recognition performance, multi-scale Gabor orientation Weber local descriptors are proposed in view of the rich line features of palmprint images. First, multi-scale Gabor filter is used to filter the palmprint image to generate multi-scale energy maps and orientation maps. Then, the differential excitation is calculated based on energy maps. Finally, the histogram features are constructed based on multi-scale differential excitation maps and orientation maps, and the feature vectors from different scales are then concatenated to produce the final feature set of a palmprint image. The experiments on PolyU, PolyU Multi-spectral and CASIA palmprint databases show that the proposed method can achieve higher identification rate and lower equal error rate compared with some existing palmprint recognition methods.
  • Jul. 16, 2021
  • Laser & Optoelectronics Progress
  • Vol.58 Issue, 16 1610018 (2021)
  • DOI:10.3788/LOP202158.1610018
Image registration of the dual-channel spaceborne hyperspectral imager with motion compensation
Zhao Huijie, Zhang Xiaoyuan, Jia Guorui, Qiu Xianfei, and Zhai Liang
Latest generation of dual-channel spaceborne hyperspectral imager based on visible near infrared (VNIR) and short wave infrared (SWIR) uses the field slitter to separate VNIR and SWIR channels into several sub-fields, and each sub-field image has different ground area at the same time. When using motion compensation technology to improve signal to noise ratio of the instrument, the observation angles of each sub-field are different, which leads to more complicated mismatch of images and make it impossible to get the continuous VNIR-SWIR spectrum of the ground pixel. The rule of image distortion and dual-channel mismatch quantitatively was analyzed by Rigorous imaging model, and the registration scheme by using geometric orientation of each sub-field separately as well as the phase correlation method was proposed on this basis. Verification based on dual-channel spaceborne hyperspectral simulated data of Dongtianshan under motion compensation was performed. The result shows that registration accuracy of traditional scheme based on correlation of images reaches 3.9 pixel, which means the continuous VNIR-SWIR spectrum of the ground pixel is still unavailable. The registration accuracy of the scheme proposed by this paper reaches 0.3 pixel, and the reflectance spectrum overlap ratio error of the overlapping bands of VNIR and SWIR reduces from 41.5% to 1.2%.
  • Jul. 15, 2021
  • Infrared and Laser Engineering
  • Vol.50 Issue, 3 20211022 (2021)
  • DOI:10.3788/IRLA20211022
Siamese networks tracking algorithm integrating channel-interconnection-spatial attention
Cui Zhoujuan, An Junshe, and Cui Tianshu
The tracking algorithms based on the Siamese networks show great potential in terms of tracking accuracy and speed. However, it is still challenging to adapt the offline trained model to online tracking. In order to improve the feature extraction and discrimination ability of the algorithm in complex scenes, a Siamese network real-time tracking algorithm that combines channel, interconnection and spatial attention mechanisms was proposed. First a Siamese tracking framework with a deep convolutional network VGG-Net-16 as the backbone network was built to increase feature extraction capabilities; then the channel-interconnection-spatial attention module was integrated to enhance the adaptability and discrimination capabilities of the model; then the multi-layer response maps were weighted and fused to obtain more accurate tracking results; and finally the large-scale datasets were used to train the end-to-end network, and tracking test on the benchmark OTB-2015 was completed. The experimental results show that compared with the current mainstream algorithms, the proposed algorithm is more robust and better adapt to complex scenes such as target appearance changes, similar distractors, and occlusion. On the NVIDIA RTX 2060 GPU, the average tracking speed reaches 37FPS, which meets real-time requirements.
  • Jul. 15, 2021
  • Infrared and Laser Engineering
  • Vol.50 Issue, 3 20200148 (2021)
  • DOI:10.3788/IRLA20200148
Non-uniform strong noise removal method for non-cooperative mine target image
Hong Hanyu, Wu Shikang, Shi Yu, Wu Jinmeng, and Sun Chunsheng
The detection of mine targets will be interfered by the underwater non-uniform strong noise (organic matter, suspended particles, etc). To solve this problem, a novel denoising method was proposed. Firstly, the local edge preserving filtering algorithm was optimized and the local edge preserving filtering based on edge perception constraint was proposed. A spatially adaptive edge perception constraint regularization term was introduced into the model to better represent the edges and details of the image, so that the edge-preserving and smoothing property could be better. Secondly, the multi-scale strategy was used to solve the heterogeneity of strong noise, the optimized model was iteratively applied to the noise removal results of each scale to generate multi-scale decomposition, and the denoising scale was gradually increaseed in the process of multi-scale decomposition. The noise of different scales was gradually separated from the denoising results of the previous scale. The experimental results show that, compared with other classical denoising methods, the proposed algorithm can better remove the underwater non-uniform strong noise while retaining the mine target information, which also has a certain guiding significance for real-time mine operation.
  • Jul. 15, 2021
  • Infrared and Laser Engineering
  • Vol.50 Issue, 3 20200344 (2021)
  • DOI:10.3788/IRLA20200344
An improved semi-supervised transfer learning method for infrared object detection neural network
Li Weipeng, Yang Xiaogang, Li Chuanxiang, Lu Ruitao, and Huang Pan
In view of the infrared datasets which has limited scale and few labeled samples, a semi-supervised transfer learning method was proposed for the training of infrared object detection neural network. It aimed at improving the training efficiency and generalization ability of object detection neural networks on infrared datasets with limited scale, and increasing the adaptability of deep learning models in scenarios with few training samples such as infrared object detection. Firstly, the ability of unlabeled samples in improving model generalization and suppressing overfitting under few labeled samples was described. Then, the process of semi-supervised transfer learning for infrared object detection neural network was proposed: a pre-trained model was trained on large scale RGB dataset, and next it was fine-tuned using a few labeled and unlabeled IR images. Moreover, a pseudo-supervised loss function with feature similarity weighting was proposed, where the predictions from same batch was used as labels to each other, thus making full use of the feature distribution of similar objects in unlabeled images. To reduce the computation of semi supervised learning, the pseudo-supervised loss of object was limited on the objects within the neighborhood of its feature vector. Experimental results show that the test accuracy of object detection neural network trained by proposed method is higher than that trained by supervised transfer learning, it achieves an improvement of 1.1% on Faster R-CNN and a significant improvement of 4.8% on YOLO-v3, which verifies the effectiveness of the proposed method.
  • Jul. 15, 2021
  • Infrared and Laser Engineering
  • Vol.50 Issue, 3 20200511 (2021)
  • DOI:10.3788/IRLA20200511
RGBT dual-modal Siamese tracking network with feature fusion
Shen Yali
Infrared imaging technology has been widely used for object tracking in military, remote sensing, security and other fields. However, thermal infrared images generally suffer from low contrast and blurry targets. Therefore, it has great importance of fusing infrared images with visible images. Compared with single-modal RGB trackers, dual-modal RGBT(RGB/Thermal infrared) trackers are more robust to illumination variation and fog. In this paper, a RGBT dual-modal siamese tracking network with feature fusion was proposed. Convolutional features extracted from the visible image and infrared image were fused to improve the appearance feature discrimination. The network can use the training data for end-to-end off-line training. Experimental results on the public RGBT234 dataset demonstrate that our tracker achieves robust and persistent tracking in complex scenarios.
  • Jul. 15, 2021
  • Infrared and Laser Engineering
  • Vol.50 Issue, 3 20200459 (2021)
  • DOI:10.3788/IRLA20200459