2023

• Record 13 of
  
  Title:Gesture Estimation Method Based on Efficient MultiScale Fusion Network
  
  Author(s):Zhuang, Xiaoxuan(1); Zhang, Man(1); Ming, Jiahui(2); Gao, Sihan(3); Chen, Hongyao(4); Zhu, Jing(5)
  Source: 2023 6th International Conference on Artificial Intelligence and Big Data, ICAIBD 2023  Volume: null  Issue: null  Article Number: null  DOI: 10.1109/ICAIBD57115.2023.10206079  Published: 2023  
  Abstract:Gesture estimation is a vital but challenge task for human-computer interaction. A desired gesture estimation method is supposed to be efficient for real-time requirement, yet still accurate. In this paper, we propose a novel network named Efficient Multi-Scale Fusion Net(EMF-Net), which effectively trades off between latency and accuracy. The cores of EMF-Net are two subnetworks, a efficient feature extraction network and a multi-scale fusion network. By presenting an improved feature extraction network to reduce latency of the model, and creating a multi-scale fusion scheme to alleviate the weak spatial generalization ability, the two subnetworks greatly improve accuracy while maintaining efficiency. Experimental results demonstrate our method attains superior accuracy with comparable computational burden. © 2023 IEEE.
  Accession Number: 20233614680589
• Record 14 of
  
  Title:150 Gbps multi-wavelength FSO transmission with 25-GHz ITU-T grid in the mid-infrared region
  
  Author(s):Su, Yulong(1); Meng, Jiacheng(2,3); Wei, Tingting(2,3); Xie, Zhuang(2,3); Jia, Shuaiwei(2,3); Tian, Wenlong(1); Zhu, Jiangfeng(1); Wang, Wei(2,3)
  Source: Optics Express  Volume: 31  Issue: 9  Article Number: null  DOI: 10.1364/OE.487668  Published: April 24, 2023  
  Abstract:The 3∼5 µm mid-infrared (mid-IR) light has several exceptional benefits in the case of adverse atmospheric conditions compared to the 1.5 µm band, so it is a promising candidate for optical carriers for free-space communication (FSO) through atmospheric channels. However, the transmission capacity in the mid-IR band is constrained in the lower range due to the immaturity of its devices. In this work, to replicate the 1.5 µm band dense wavelength division multiplexing (DWDM) technology to the 3 µm band for high-capacity transmission, we demonstrate a 12-channel 150 Gbps FSO transmission in the 3 µm band based on our developed mid-IR transmitter and receiver modules. These modules enable wavelength conversion between the 1.5 µm and 3 µm bands based on the effect of difference-frequency generation (DFG). The mid-IR transmitter effectively generates up to 12 optical channels ranging from 3.5768 µm to 3.5885 µm with a power of 6.6 dBm, and each channel carries 12.5 Gbps binary phase shift keying (BPSK) modulated data. The mid-IR receiver regenerates the 1.5 µm band DWDM signal with a power of -32.1 dBm. Relevant results of regenerated signal demodulation have been collected in detail, including bit error ratio (BER), constellation diagram, and eye diagram. The power penalties of the 6th to 8th channels selected from the regenerated signal are lower than 2.2 dB compared with back-to-back (BTB) DWDM signal at a bit error ratio (BER) of 1E-6, and other channels can also achieve good transmission quality. It is expected to further push the data capacity to the terabit-per-second level by adding more 1.5 µm band laser sources and using wider-bandwidth chirped nonlinear crystals. © 2023 Optica Publishing Group.
  Accession Number: 20231914062482
• Record 15 of
  
  Title:Adaptive Style Modulation for Artistic Style Transfer
  
  Author(s):Zhang, Yipeng(1,2,3); Hu, Bingliang(1,2); Huang, Yingying(1,2,3); Gao, Chi(1,2,3); Wang, Quan(1,2)
  Source: Neural Processing Letters  Volume: 55  Issue: 5  Article Number: null  DOI: 10.1007/s11063-022-11135-7  Published: October 2023  
  Abstract:Arbitrary-style-per-model (ASPM) style transfer algorithms transfer arbitrary styles based on a single model. Statistics-based learning algorithms of ASPM, represented by adaptive instance normalization (AdaIN), conduct instance normalization and then perform an affine transformation on target features. These algorithms are computationally efficient and easy to embed in convolutional neural networks. Consequently, they are widely used in image synthesis tasks to control the style of the resulting images. However, the style of stylized images may be a combination of content and stylized images, which suggests that these methods do not transform styles accurately. In this work, we rethink the function of AdaIN in controlling style. We show that the role of AdaIN is to (1) give each input content image a specific optimization target, (2) dynamically set cross-channel correlations, and (3) act as a feature selector after combining it with an activation function. Accordingly, we propose adaptive style modulation (AdaSM), which fully leverages the three roles mentioned above and thereby enables more precise control of global style. Experimental results show that AdaSM provides superior style controllability, alleviates the style blending problem, and outperforms state-of-the-art methods in artistic style transfer tasks. © 2022, The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature.
  Accession Number: 20230113327717
• Record 16 of
  
  Title:Research on two assembling methods of 650 mm SiC lightweight mirror
  
  Author(s):Wang Tao, Ren(1); Peng, Wang(1)
  Source: Proceedings of SPIE - The International Society for Optical Engineering  Volume: 12976  Issue: null  Article Number: 129761R  DOI: 10.1117/12.3009485  Published: 2023  
  Abstract:The primary mirror system is the key component of the high-precision optical system, and the surface accuracy of the primary mirror determines the imaging quality of the whole system. When the lightweight mirror is affected by its own weight, support mode, load and temperature, the surface shape accuracy of the primary mirror will be changed, and the assembly mode of the primary mirror is an important factor to ensure the surface shape accuracy of the primary mirror. There are usually three types of assembly modes for the primary mirror: peripheral support, central support and back support. In this paper, the supporting mode of 650mm lightweight mirror is studied, and the surface shape of the mirror is analyzed under two supporting modes (center support and back support) when the elevation of the mirror is between 0 ° and 60 °, which is affected by its own weight. Firstly, the primary mirror is modeled based on CAD; then, the CAE mesh is generated according to the CAD model with different support modes, and the structural deformation of the primary mirror under different working conditions is obtained by solving the structure of the primary mirror under different working conditions; finally, the mesh data of CAE analysis is extracted, and the mesh data is subjected to polynomial fitting to obtain the deformation of the secondary mirror in different working conditions and States. Based on the method of simulation analysis, this paper adopts a variety of simulation tools, including CAD, CAE, multidisciplinary fitting analysis software and so on. The deformation of the primary mirror under different working conditions and different support modes is obtained. When the 650 mm lightweight mirror is supported on the back, the surface shape change does not exceed RMS 0. 01 λ at different elevation angles. For the center support, the surface shape change is RMS 0. 004 λ to RMS 0. 12 λ in the elevation angle range of 0 ° to 60 °. This supporting mode has a great influence on the surface shape. Based on the simulation analysis data, the back support is used for the assembly of the 650 mm lightweight mirror, and the center support is not suitable for the assembly of the mirror. © COPYRIGHT SPIE.
  Accession Number: 20240315402124
• Record 17 of
  
  Title:MinimalGAN: diverse medical image synthesis for data augmentation using minimal training data
  
  Author(s):Zhang, Yipeng(1,2,3); Wang, Quan(1,2); Hu, Bingliang(1,2)
  Source: Applied Intelligence  Volume: 53  Issue: 4  Article Number: null  DOI: 10.1007/s10489-022-03609-x  Published: February 2023  
  Abstract:Image synthesis techniques have limited application in the medical field due to unsatisfactory authenticity and precision. Additionally, synthesizing diverse outputs is challenging when the training data are insufficient, as in many medical datasets. In this work, we propose an image-to-image network named the Minimal Generative Adversarial Network (MinimalGAN), to synthesize annotated, accurate, and diverse medical images with minimal training data. The primary concept is to make full use of the internal information of the image and decouple the style from the content by separating them in the self-coding process. After that, the generator is compelled to concentrate on content detail and style separately to synthesize diverse and high-precision images. The proposed MinimalGAN includes two image synthesis techniques; the first is style transfer. We synthesized a stylized retinal fundus dataset. The style transfer deception rate is much higher than that of traditional style transfer methods. The blood vessel segmentation performance increased when only using synthetic data. The other image synthesis technique is target variation. Unlike the traditional translation, rotation, and scaling on the whole image, this approach only performs the above operations on the segmented target being annotated. Experiments demonstrate that segmentation performance improved after utilizing synthetic data. © 2022, The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature.
  Accession Number: 20222312202688
• Record 18 of
  
  Title:Quadri-Wave Lateral Shearing Interference Wavefront Reconstruction Based on Path Guidance
  
  Author(s):Min, Xingzhi(1,2,3); Duan, Yaxuan(1,3); Wang, Zhengzhou(1,3); Chen, Xiaoyi(1,3); Tang, Zhiyuan(1,3); Wang, Pu(1,3); Fan, Yao(1,3)
  Source: Zhongguo Jiguang/Chinese Journal of Lasers  Volume: 50  Issue: 18  Article Number: 1804003  DOI: 10.3788/CJL221344  Published: September 25, 2023  
  Abstract:Objective The accuracy of a quadri-wave lateral shearing interferometer is directly affected by the accuracy of the wavefront reconstruction. Traditional wavefront-reconstruction methods include modal and zonal methods. The modal method expands the wavefront into a set of primary functions to be measured, then fits the coefficients corresponding to the primary functions to reconstruct the measured wavefront. The zonal method discretizes the measured wavefront to establish a mapping relationship between the measured and differential wavefronts for reconstruction. Alternatively, the wavefront can be reconstructed by direct integration in the shearing direction. However, the modal method always uses finite terms to fit the measured wavefront, which directly ignores high-frequency information, reducing the estimated accuracy of the quadri-wave lateral shearing interferometer. The zonal method has a high spatial resolution, but the noise error accumulates along the integrated path during the reconstruction process, forming noise lines, thus, affecting the accuracy of the reconstructed wavefront. To solve this problem, a quadri-wave lateral shearing interferometric wavefront reconstruction method is proposed based on path guidance, which has both high accuracy and spatial resolution. Methods In this study, a theoretical analysis of the drawbacks of noise error accumulation in wavefront reconstruction using the zonal method without integral-path guidance under noisy environments is carried out. An integral-path evaluation-map model is established based on the deviation of differential phase derivatives, and a flowchart of the wavefront reconstruction algorithm is provided based on integral-path guidance. The proposed method consists of two steps. First, the evaluation model of the differentialphase-derivative deviation is used to count the variational characteristics of the differential phase, identify the noise error, and generate an integral path to avoid noise error. Second, the generated path is used to guide the wavefront reconstruction integral of the Southwell model. Using theoretical simulations, the proposed method could effectively prevent the propagation and accumulation of noise errors compared to the zonal method without integral-path guidance under noisy environments for different signal-to-noise ratios (SNRs). In addition, a verification device having a pure-phase liquid-crystal spatial light modulator (SLM) was set up to experimentally verify the effectiveness of the proposed method. The experimental results of the proposed method were also compared with those of the zonal method without integral-path guidance. Results and Discussions In the simulation, interferograms with a sinusoidal phase distribution are generated (Fig. 4). When the SNR increases from 10 dB to 50 dB, the root-mean-square (RMS) between the wavefront reconstructed by the zonal method without integral-path guidance and the theoretical wavefront decreases from 0.0152λ to 0.0094λ. However, the RMS between the wavefront reconstructed by the proposed method and the theoretical wavefront decreases from 0.0139λ to 0.0041λ. Moreover, the proposed method reduces the RMS of the reconstructed and theoretical wavefronts by a maximum of 55.6% compared to the zonal method without integral-path guidance (Fig. 7). Thus, the proposed method is more robust than the zonal method without integral-path guidance under the Gaussian noise environment with different SNRs (Fig. 6). In the experiment, we measure the random phase generated by the spatial light modulator using the proposed method and zonal method without integral-path guidance (Fig. 10). The results show that the PV value (peak-valley value) of the wavefront reconstructed by the proposed method is 0.7283λ, whereas that of the wavefront reconstructed by the zonal method without integral-path guidance is 2.966λ. The deviation between the PV value of the wavefront reconstructed by the proposed method and that of the theoretical wavefront is 1.6943λ, which is less than the deviation between the wavefront PV value reconstructed by the zonal method without integral-path guidance and the theoretical wavefront PV value (Fig. 13). In addition, the RMS between the wavefront reconstructed by the proposed method and the theoretical wavefront is reduced by 39.7% compared with the RMS between the wavefront reconstructed by the zonal method without integral-path guidance and the theoretical wavefront. In addition, the zonal method without integral-path guidance is used to reconstruct the wavefront, which propagates and accumulates noise points along the shearing direction by forming noise lines. However, the proposed method prevents the propagation of noise points and improves wavefront reconstruction accuracy. Conclusions This paper proposes a quadri-wave lateral shearing interference wavefront reconstruction method based on integral-path guidance. The effectiveness of the proposed method is verified through theoretical simulations and experiments. The theoretical simulation results show that the proposed method prevents the propagation of noise errors and improves the wavefront reconstruction accuracy compared to the zonal method without integral-path guidance under noisy environments with different SNRs. The RMS between the reconstructed wavefronts of the proposed method and the theoretical wavefront is smaller than that between the zonal method without integral-path guidance and the theoretical wavefront under the same conditions. Moreover, the experimental results show that the proposed method can effectively prevent the propagation and accumulation of differential phase noise points when measuring the random phase generated by the pure-phase liquid crystal spatial light modulator and reconstructing the wavefront of the random phase. However, the wavefront of the random phase reconstructed using the zonal method without integral-path guidance cannot be accurately reconstructed because of the noise line generated by the accumulation of noise. Therefore, the proposed method has higher accuracy and robustness than the zonal method without integral-path guidance in reconstructing the wavefront in a noisy environment. © 2023 Science Press. All rights reserved.
  Accession Number: 20234114851695
• Record 19 of
  
  Title:Preparation and imaging properties of coherent chalcogenide glass fiber bundles with large planar array for far-infrared transmission (invited)
  
  Author(s):Liu, Xiaogang(1,2); Xu, Yantao(2); Guo, Haitao(2); Yan, Xingtao(2); Kong, Depeng(2); Shen, Xiaoming(1); Chang, Yanjie(2,3); Zhang, Hao(2,3)
  Source: Hongwai yu Jiguang Gongcheng/Infrared and Laser Engineering  Volume: 52  Issue: 5  Article Number: 20230110  DOI: 10.3788/IRLA20230110  Published: May 2023  
  Abstract:Objective The 8-10 μm far-infrared spectrum is in the infrared radiation band at natural temperatures and covers the characteristic "fingerprint spectrum" of many molecules, so it has important applications in the military, medical and environmental monitoring fields. Infrared coherent fiber bundles which can realize the flexible transmission of infrared image are the basic components for assembling various infrared optical systems, and they can be used in the narrow space, high-intensity electric or magnetic field in particular. The main types of far infrared fibers mainly include crystal fiber, hollow fiber, photonic crystal fiber and Te- based chalcogenide glass fiber. Among them, Te-based fiber is an excellent far-infrared transmission material due to its wide transmission band, stable thermal, chemical properties, which means it is especially suitable for the preparation of coherent optical fiber bundles with large array. Until now, a series of components such as Ge-As-Se-Te, GeTe-AgI, Ga-Ge-Te, Ge-Te-I and As-Se-Te have been studied. However, the optical loss of Te-base fiber is still higher at present, which limits the transmission distance of infrared signal and the resolution of the infrared bundles. Therefore, it is necessary to study the purification technology for optimizing the optical loss. Methods High purity raw materials of As, Se and Te were purified by multi-distillation purification technique and the content of O element was examined by EPMA. As-Se-Te chalcogenide glass was chosen and melted by different preparation process and their infrared transmission spectra were measured by FTIR. The optical fiber was drawn by the rod-in-tube method. The drawing temperature was 240 ℃ with the accuracy of ±0.2 ℃, and the drawing speed was about 10 m/min. The coherent fiber bundle was prepared by ribbon-stacking technique. The end face was observed by microscope. Infrared image was detected by home-made optical system and mercury cadmium telluride detector was used (Fig.2). Results and Discussions The oxygen content of As, Se, Te raw materials decreased from 1.3 at%, 0.46 at% and 0.48 at% in raw materials to 0 at% (undetected), 0.06 at% and 0.15 at% in purified materials respectively, indicating that the distillation process was effective (Tab.1). The transition temperature Tg is 137.5 ℃ for core material and 139.1 ℃ for clad material (Fig.3), which are very close and match well. No obvious crystallization peak was observed in the test temperature range, indicating that the core and clad glass are suitable for fiber drawing. Smooth spectrum was obtained in the sample of aluminum as a deaerator (Fig.4). The optical fiber with an outer diameter of 100 μm was obtained. Its bending radius is less than 5 mm, and the baseline of the optical loss is about 0.2 dB/m in the far infrared range (Fig.5). Finally, the coherent fiber bundle with 22.5 thousand pixels and close-packed arrangement was prepared. The total fracture rate is less than 3 and there are none black or dark pixels in the center region of the bundle. The bundle transmits infrared beam uniformly and the image of the infrared target is clear and distortionless, which indicates that the comprehensive properties of the bundle are satisfactory (Fig.6). Conclusions Far-infrared fiber bundles was prepared and measured. In order to eliminate impurities, As-Se-Te chalcogenide glass was chosen and the high purity raw materials of As, Se and Te were purified. As-Se-Te glasses were melting by different preparation process and their infrared transmission spectra were measured and analyzed. The results show that excellent thermal and far-infrared transmitting performance can be obtained in the sample of Al as deoxidizer process. The optical fiber was drawn with an outer diameter of 100 μm, bending radius of less than 5 mm, optical loss of 0.2 dB/m. The coherent fiber bundle was prepared by ribbon-stacking technique. It has 22.5 thousand pixels and the total fracture rate is less than 3. The infrared target imaging was distortionless and showed fine temperature resolution, demonstrating that the bundles can be widely used in infrared imaging systems. © 2023 Chinese Society of Astronautics. All rights reserved.
  Accession Number: 20232814375583
• Record 20 of
  
  Title:Light Gap Bullets in Defocusing Media with Optical Lattices
  
  Author(s):Chen, Zhiming(1,2); Wu, Zexing(2); Zeng, Jianhua(1,3)
  Source: SSRN  Volume: null  Issue: null  Article Number: null  DOI: 10.2139/ssrn.4471674  Published: June 7, 2023  
  Abstract:Searching for three-dimensional spatiotemporal solitons (also known as light/optical bullets) has recently attracted keen theoretical and experimental interests in nonlinear physics. Currently, optical lattices of diverse kinds have been introduced to the stabilization of light bullets, while the investigation for the light bullets of gap type-nonlinear localized modes within the finite gap of the underlying linear Bloch spectrum-is lacking. Herein, we address the formation and stabilization properties of such light gap bullets in periodic media with defocusing nonlinearity, theoretically and in numerical ways. The periodic media are based on two-dimensional periodic standing waves created in a coherent three-level atomic system which is driven to the regime of electromagnetically induced transparency, which in principle can also be replaced by photonic crystals in optics or optical lattices in ground-state ultracold atoms system. The temporal dispersion term is tuned to normal (positive) group velocity dispersion so that to launch the light gap bullets under self-repulsive nonlinearity; two types of such light gap bullets constructed as three-dimensional gap solitons and vortices with topological charge m=1 within the first finite gap are reported and found to be robustly stable in the existence domains. On account of the light bullets were previously limited to the semi-infinite gap of periodic media and continuous nonlinear physical systems, the light gap bullets reported here thus supplement the missing type of three-dimensional spatiotemporal localized modes in periodic media which exhibit finite band gaps. © 2023, The Authors. All rights reserved.
  Accession Number: 20230182223
• Record 21 of
  
  Title:Reconstruction algorithm using 2N+1 raw images for structured illumination microscopy
  
  Author(s):Fang, Xiang(1); Wen, Kai(1); An, Sha(1); Zheng, Juanjuan(1,2); Li, Jianlang(1); Zalevsky, Zeev(3); Gao, Peng(1)
  Source: Journal of the Optical Society of America A: Optics and Image Science, and Vision  Volume: 40  Issue: 4  Article Number: null  DOI: 10.1364/JOSAA.483884  Published: April 2023  
  Abstract:This paper presents a structured illumination microscopy (SIM) reconstruction algorithm that allows the reconstruction of super-resolved images with 2N + 1 raw intensity images, with N being the number of structured illumination directions used. The intensity images are recorded after using a 2D grating for the projection fringe and a spatial light modulator to select two orthogonal fringe orientations and perform phase shifting. Super-resolution images can be reconstructed from the five intensity images, enhancing the imaging speed and reducing the photobleaching by 17%, compared to conventional two-direction and three-step phase-shifting SIM. We believe the proposed technique will be further developed and widely applied in many fields. © 2023 Optica Publishing Group.
  Accession Number: 20231513870228
• Record 22 of
  
  Title:Multiband camouflage design with thermal management
  
  Author(s):Huang, Lehong(1,2,3,4); Li, Haochuan(1); Li, Zhiguo(1,3,4); Zhang, Wenbo(1,2,3); Ma, Caiwen(1,3,4); Zhang, Chunmin(2); Wei, Yuxuan(1,3,4); Zhou, Liang(1,4); Li, Xun(1); Cheng, Zhiyuan(1,3,4); Guo, Xiaohui(1); Guo, Shiping(2)
  Source: Photonics Research  Volume: 11  Issue: 5  Article Number: null  DOI: 10.1364/PRJ.484448  Published: May 2023  
  Abstract:Although the effective "stealth" of space vehicles is important, current camouflage designs are inadequate in meeting all application requirements. Here, a multilayer wavelength-selective emitter is demonstrated. It can realize visible light and dual-band mid-infrared camouflage with thermal control management in two application scenarios, with better effect and stronger radiation cooling capability, which can significantly improve the stealth and survivability of space vehicles in different environments. The selective emitter demonstrated in this paper has the advantages of simple structure, scalability, and ease of large-area fabrication, and has made a major breakthrough in driving multiband stealth technology from simulation research to physical verification and even practical application. © 2023 Chinese Laser Press.
  Accession Number: 20232214167431
• Record 23 of
  
  Title:Middle and near ultraviolet spectrograph of the Scientific Experimental system in Near SpacE (SENSE)
  
  Author(s):Sun, Xin(1,2); Shi, Dalian(1,2); Chen, Zhen(1,2); Li, Ran(1,2); Cao, Weiwei(1,2); Zhu, Jun(1,2); Bai, Yonglin(1,2); Wang, Le(3); He, Fei(4)
  Source: Earth and Planetary Physics  Volume: 7  Issue: 6  Article Number: null  DOI: 10.26464/epp2023081  Published: 2023  
  Abstract:The Scientific Experimental system in Near SpacE (SENSE) consists of different types of instruments that will be installed on a balloon-based platform to characterize near-space environmental parameters. As one of the main scientific payloads, the middle and near ultraviolet spectrograph (MN-UVS) will provide full spectra coverage from middle ultraviolet (MUV, 200−300 nm) to near ultraviolet (NUV, 300−400 nm) with a spectral resolution of 2 nm. Its primary mission is to acquire data regarding the UV radiation background of the upper atmosphere. The MN-UVS is made up of six primary components: a fore-optical module, an imaging grating module, a UV intensified focal plane module, a titanium alloy frame, a spectrometer control module, and a data processing module. This paper presents in detail the engineering design of each functional unit of the MN-UVS, as well as the instrument’s radiometric calibration, wavelength calibration, impact test, and low-pressure discharge test. Furthermore, we are able to report ground test and flight test results of high quality, showing that the MN-UVS has a promising future in upcoming near-space applications. © 2023 by Earth and Planetary Physics.
  Accession Number: 20234715073538
• Record 24 of
  
  Title:Adaptive convolution kernel network for change detection in hyperspectral images
  
  Author(s):Liu, Song(1,2); Li, Haiwei(1); Chen, Junyu(1,2); Li, Siyuan(1); Song, Liyao(3); Zhang, Geng(1); Hu, Bingliang(1)
  Source: Applied Optics  Volume: 62  Issue: 8  Article Number: null  DOI: 10.1364/AO.479955  Published: March 10, 2023  
  Abstract:Feature extraction is a key step in hyperspectral image change detection. However, many targets with great various sizes, such as narrow paths, wide rivers, and large tracts of cultivated land, can appear in a satellite remote sensing image at the same time, which will increase the difficulty of feature extraction. In addition, the phenomenon that the number of changed pixels is much less than unchanged pixels will lead to class imbalance and affect the accuracy of change detection. To address the above issues, based on the U-Net model, we propose an adaptive convolution kernel structure to replace the original convolution operations and design a weight loss function in the training stage. The adaptive convolution kernel contains two various kernel sizes and can automatically generate their corresponding weight feature map during training. Each output pixel obtains the corresponding convolution kernel combination according to the weight. This structure of automatically selecting the size of the convolution kernel can effectively adapt to different sizes of targets and extract multi-scale spatial features. The modified cross-entropy loss function solves the problem of class imbalance by increasing the weight of changed pixels. Study results on four datasets indicate that the proposed method performs better than most existing methods. © 2023 Optica Publishing Group.
  Accession Number: 20231513879442