Image Simulation Method of Infrared UAV Based on Image Derivation
Zhang Yu, Zhang Yan, Shi Zhiguang, Zhang Jinghua, Liu Di, Suo Yuchang, Shi Xiaoran, and Du Jinming
Monitoring the infrared video of the unmanned aerial vehicle (UAV) group is a new hot spot in the security and military fields. Due to the difficulty of acquiring UAV images in complex backgrounds, and the number of images cannot meet the requirements of model training and verification of related algorithms, an image simulation method of infrared UAV based on image derivation is proposed. This method is used to simulate infrared UAVs. This method is used to mix the infrared UAV template image and the infrared background image to generate a large number of UAV target images in different backgrounds. Aiming at the problems of image mixing technology being severely affected by background noise, blurring of drone target edges, and low harmony of synthetic images, an unsupervised generative confrontation network is used to generate a gray-scale constrained image with a high degree of harmony. The target gradient image is used as a joint constraint to solve the Gaussian-Poisson equation, and a mixed image with high consistency with the real image characteristics is obtained. The experimental results show that the mixed image generated by the proposed method has high image harmony and visual authenticity, which shows that the obtained image as an extended sample can effectively improve the performance of the machine learning algorithm.
  • Dec. 29, 2021
  • Acta Optica Sinica
  • Vol.42 Issue, 2 0210003 (2022)
  • DOI:10.3788/AOS202242.0210003
Variational Mode Decomposition and Wavelet Threshold Function De-Noising for Second Harmonics
Zhang Ruilin, and Tu Xinghua
A second harmonic de-noising method based on variational mode decomposition and wavelet threshold function is proposed to solve the problem of external noise interference in second harmonic spectra during gas concentration measurement by tunable diode laser absorption spectroscopy (TDLAS). In this paper, we decompose the noisy second harmonic signal to get the useful intrinsic mode functions (IMFs) and reconstruct them. Then, we conduct the de-noising process for the reconstructed signal with the wavelet threshold function. The selection of the optimal balance parameter in the variational mode decomposition is discussed, and the proportional relationship of the optimal balance parameter with the noise in the noisy signal is obtained. Better noise suppression is achieved by changing the threshold function of wavelet transform and thereby altering the high-frequency wavelet coefficients. The de-noising results of actual measurement curves show that the proposed de-noising method can effectively suppress the noise and extract the useful second harmonic signal in the case of a poor signal-to-noise ratio.
  • Dec. 29, 2021
  • Acta Optica Sinica
  • Vol.42 Issue, 2 0210001 (2022)
  • DOI:10.3788/AOS202242.0210001
Segmentation Method of Broken Ore Image Based on Improved HED Network Model
Gu Qinghua, Wei Fawen, Guo Mengli, Jiang Song, and Ruan Shunling
The particle size of ore is an important reference to judge the crushing effect of crusher, and image segmentation is the key step of ore particle size detection. To solve the problems of image segmentation inaccuracies caused by complex shape, adhesion and stacking of broken ore, and serious image noise, a broken ore image segmentation method based on improved HED (Holistically-Nested Edge Detection) network model is proposed. First, the bilateral filtering pre-processing operation is carried out on the collected ore image to reduce the influence of noise on segmentation. Second, the residual deformable convolution block is used to replace the ordinary convolution block to enhance the feature extraction ability of the model for ores of different sizes and shapes, and the void convolution is used to replace the original pooling layer to expand the receptive field and retain the global information of ores. Finally, the HED network framework with a bottom-short connection structure is used for feature extraction of ore, and the extracted features are combined with low-level detail information to reduce the problem of undersegmentation of cohesive and stacked ore particles.
  • Dec. 23, 2021
  • Laser & Optoelectronics Progress
  • Vol.59 Issue, 2 0210020 (2022)
  • DOI:10.3788/LOP202259.0210020
Optimization Method for Infrared Eye Movement Image Segmentation
Li Chang, Liu Yu, and Sun Jinglin
When the eye tracker is collecting infrared eye movement data, due to the rapid movement of the subject’s eyeballs or the inability to keep relatively still with the instrument, some of the collected eye area images are defocused and blurred. This paper proposes a semantic segmentation optimization system, which is called super real-time semantic segmentation network (S-RITnet). First, a pixel-level annotation data set with a 4∶1∶1 ratio of images in the training set, validation set, and test set is created. Then, the enhance super-resolution generative adversarial network and contrast-limited adaptive histogram enhancement algorithm are used to repair the blurred eye area data set image. Finally, based on real-time semantic segmentation net and the autonomous data set (including the repair data set), perform network training to realize the semantic segmentation of the eye area image and evaluate the obtained segmentation module. The experimental results show that the optimization scheme can effectively optimize the quality of eye area images. Compared with the low-quality eye images training module, the mean intersection over union and F1-score evaluation of S-RITnet increased by 0.0247 and 0.024 respectively.
  • Dec. 23, 2021
  • Laser & Optoelectronics Progress
  • Vol.59 Issue, 2 0210016 (2022)
  • DOI:10.3788/LOP202259.0210016
Recognition Algorithm of Dangerous Goods in Security Inspection Based on Multi-Layer Attention Mechanism
Wang Wen, Zhou Yatong, Shi Baojun, He Hao, and Zhang Jianwei
Aiming at the problems that the data set used by the existing recognition algorithms is too simple, the recognition accuracy of dangerous goods in security inspection images in real scenes is low, and it is easy to lead to false detection and missed detection, we propose a class-balanced hierarchical refinement algorithm based on penetration hypothesis, which combines multi-layer channel attention mechanism and space attention mechanism. First, based on the hierarchical modeling of security image, channel attention mechanism is added to the feature map to give different weight to different channel features. Then, spatial attention mechanism is added to give different weight to the unique color features of security image in space. Finally, the residual network is used to add double attention mechanism to different layers of security image for ablation experiment. The experimental results show that after adding double attention mechanism to the fixed two layers at the same time, the network can significantly improve the identification accuracy of dangerous goods in security inspection, and verify the effectiveness and robustness of the multi-layer attention mechanism algorithm.
  • Dec. 23, 2021
  • Laser & Optoelectronics Progress
  • Vol.59 Issue, 2 0210011 (2022)
  • DOI:10.3788/LOP202259.0210011
A Defogging Method Containing Images of Sky
Guo Tongying, Li Na, Sun Liangliang, and Wang Haichen
To address the poor performance of most defogging algorithms on images with large areas of sky, an improved dark channel apriori defogging method is proposed. First, segment the sky area according to the image gradient information, and based on the sky area segmentation, combine the high brightness and smoothness of the atmospheric light reference pixels to set the discriminant formula and reasonably estimate atmospheric light values. Second, a piecewise linear function is used to dynamically modify the adjustable parameters in response to the value of the dark channel to solve the local shadow caused by excessive defogging. Then, the transmittance estimated by the bright channel model and improved dark channel a priori model are fused and guided filtering is used for edge optimization. Finally, the defogging image is obtained by combining brightness compensation and contrast stretching using the atmospheric scattering model. The experimental results show that the improved method effectively reduces image distortion, improves image contrast and details, and has advantages in preserving the visual authenticity of the sky area.
  • Dec. 23, 2021
  • Laser & Optoelectronics Progress
  • Vol.59 Issue, 2 0210009 (2022)
  • DOI:10.3788/LOP202259.0210009
Defogging Algorithm Based on Image Features and Wavelet Transform
He Lifeng, Yuan Pu, Zhou Guangbin, Su Liangliang, and Lu Bofan
Aiming at the problems of halo artifact, dark distortion and detail loss in traditional dark channel prior defogging algorithms, a defogging algorithm based on image features and wavelet transform is proposed. First, The gray-level co-occurrence matrix method is introduced to obtain the complexity of image texture features as a constraint condition,and the problem of false texture and blocking effect in dark channel images is solved by use of dynamic sliding window; second, combined with the image brightness information, K-Means clustering algorithm is used to calibrate the bright and dark areas to optimize the atmospheric light value and transmittance map; finally, aiming at the problems of darkening and loss of detail features in the restored image of atmospheric scattering model, the image enhancement technology based on wavelet transform is used to improve the image contrast. The experimental results show that the proposed algorithm can recover the scence and detail features well, and performs well in peak signal to noise ratio (PSNR), structural similarity index (SSIM), and mean absolute error (MAE).
  • Dec. 23, 2021
  • Laser & Optoelectronics Progress
  • Vol.59 Issue, 2 0210021 (2022)
  • DOI:10.3788/LOP202259.0210021
Face Recognition and Age Estimation Based on Varying Illumination
Tian Huijuan, Qiao Mingtian, and Cai Minpeng
Aiming at the problem of environmental illumination in face recognition and age estimation system, a face recognition and age estimation method based on multi-task convolutional neural network under varying illumination is proposed. The recognition rate of face images and the accuracy of age estimation under varying illumination are improved by the proposed method. Retinex image enhancement algorithm in YCbCr color space is used to improve the accuracy of face recognition and age estimation, and the face recognition and age estimation experiments under 10 kinds of dimming level for three kinds of distance are carried out. Experimental results show that compared with the original images, the recognition rates of the face images obtained by the improved method are improved, and the average absolute errors of age estimation are decreased. When the dimming level is 40%, and the distance is 1, 2, and 3 m, the face recognition rates are increased by 3 percentage points, 19 percentage points, and 25 percentage points, and the average absolute error of age estimation is decreased by 1.20, 2.99 and 2.00. At the same time, it is found that the effect of face recognition and age estimation is better when the gray mean value of face image without image enhancement algorithm is more than 50.18. When it is lower than the value, it is necessary to add the image enhancement algorithm to improve the accuracy of face recognition and age estimation. After adding the image enhancement algorithm, when the gray mean value of the face image is more than 56.61, the effect of face recognition and age estimation is better, and the visual effect and image quality are better.
  • Dec. 23, 2021
  • Laser & Optoelectronics Progress
  • Vol.59 Issue, 2 0210019 (2022)
  • DOI:10.3788/LOP202259.0210019
Aerial Target Classification Algorithm Based on Double-Layer Feature Selection
Su Zhigang, and Wang Xuemeng
The classification of biological and abiotic targets in the air is an important part of bird strike control in the airport. Target classification based on trajectory information has the advantages of easy access to trajectory information and high degree of discrimination of some features, but improper feature selection will result in large classification errors of close-range trajectory samples. Aiming at this problem, an aerial target classification algorithm based on double-layer feature selection is proposed. First, fully feature extraction is performed on the three-dimensional trajectory data of dynamic targets to expand the range of feature selection. Second, the feature subset is selected through the designed two-layer feature selection algorithm, which reduces the computational complexity of the algorithm and improves the classification precision. Finally, online sequential extreme learning machine (OSELM) is used to realize the real-time classification of aerial biological and abiotic targets. Experimental results show that the proposed algorithm takes into account the accuracy and speed of classification, the classification accuracy reaches 99.7%, and the average classification time is only 1.26 ms, which meets the needs of real-time monitoring and early warning. The proposed algorithm provides a potential solution for real-time classification of air targets under airport conditions.
  • Dec. 23, 2021
  • Laser & Optoelectronics Progress
  • Vol.59 Issue, 2 0210018 (2022)
  • DOI:10.3788/LOP202259.0210018
Satellite Image Translation Method Based on Attention Residual Network
Wang Jinyu, Zhang Changgong, Yang Haitao, Feng Bodi, Li Gaoyuan, and Gao Yuge
  • Dec. 23, 2021
  • Laser & Optoelectronics Progress
  • Vol.59 Issue, 2 0210017 (2022)
  • DOI:10.3788/LOP202259.0210017