Volume 13,Issue 6,2017 Table of Contents
2017, 13(6):401-404.
DOI: https://doi.org/10.1007/s11801-017-7126-2
Abstract:
All-optical regenerators can be used to suppress amplified spontaneous emission (ASE) noise introduced by cascaded erbium doped fiber amplifiers (EDFAs) in optical fiber communication systems and lead to the improvement of optical receiver sensitivity. By introducing the Q-factor transfer function (QTF), we evaluate the Q-factor performance of degenerate four-wave mixing (DFWM) regenerators with clock pump and reveal the differences between the optimal input powers determined from the static and dynamic power tranfer function (PTF) and the QTF curves. Our simulation shows that the clock-pump regnerator is capable of improving the Q-facor and receiver sensitivity for 40 Gbit/s ASE-degraded return-to-zero on-off keying (RZ-OOK) signal by 2.58 dB and 4.2 dB, respectively.
All-optical regenerators can be used to suppress amplified spontaneous emission (ASE) noise introduced by cascaded erbium doped fiber amplifiers (EDFAs) in optical fiber communication systems and lead to the improvement of optical receiver sensitivity. By introducing the Q-factor transfer function (QTF), we evaluate the Q-factor performance of degenerate four-wave mixing (DFWM) regenerators with clock pump and reveal the differences between the optimal input powers determined from the static and dynamic power tranfer function (PTF) and the QTF curves. Our simulation shows that the clock-pump regnerator is capable of improving the Q-facor and receiver sensitivity for 40 Gbit/s ASE-degraded return-to-zero on-off keying (RZ-OOK) signal by 2.58 dB and 4.2 dB, respectively.
2017, 13(6):405-408.
DOI: https://doi.org/10.1007/s11801-017-7174-7
Abstract:
Hyper-spectral imaging spectrometer has high spatial and spectral resolution. Its radiometric calibration needs the knowledge of the sources used with high spectral resolution. In order to satisfy the requirement of source, an on-orbit radiometric calibration method is designed in this paper. This chain is based on the spectral inversion accuracy of the calibration light source. We compile the genetic algorithm progress which is used to optimize the channel design of the transfer radiometer and consider the degradation of the halogen lamp, thus realizing the high accuracy inversion of spectral curve in the whole working time. The experimental results show the average root mean squared error is 0.396%, the maximum root mean squared error is 0.448%, and the relative errors at all wavelengths are within 1% in the spectral range from 500 nm to 900 nm during 100 h operating time. The design lays a foundation for the high accuracy calibration of imaging spectrometer.
Hyper-spectral imaging spectrometer has high spatial and spectral resolution. Its radiometric calibration needs the knowledge of the sources used with high spectral resolution. In order to satisfy the requirement of source, an on-orbit radiometric calibration method is designed in this paper. This chain is based on the spectral inversion accuracy of the calibration light source. We compile the genetic algorithm progress which is used to optimize the channel design of the transfer radiometer and consider the degradation of the halogen lamp, thus realizing the high accuracy inversion of spectral curve in the whole working time. The experimental results show the average root mean squared error is 0.396%, the maximum root mean squared error is 0.448%, and the relative errors at all wavelengths are within 1% in the spectral range from 500 nm to 900 nm during 100 h operating time. The design lays a foundation for the high accuracy calibration of imaging spectrometer.
2017, 13(6):409-413.
DOI: https://doi.org/10.1007/s11801-017-7205-4
Abstract:
A tunable self-seeded multi-wavelength Brillouin-erbium fiber laser (BEFL) is proposed and demonstrated based on a few-mode fiber filter (FMFF) with varying temperature. The FMFF configuration is a section of uncoated few-mode fiber (FMF) sandwiched between two up-tapers. As the temperature varies from 25 °C to 125 °C, the transmission spectrum of FMFF moves towards the longer wavelength. The self-excited Brillouin pump is internally achieved by cascaded stimulated Brillouin scattering (SBS) in the single mode fiber (SMF). Then employing the FMFF temperature variation characteristics in the ring cavity fiber laser, the multi-wavelength of the output laser can be tuned, and the tunable range is about 8.0 nm. The generation of up to 15 Brillouin Stokes wavelengths with 16 dB optical signal-to-noise ratio (OSNR) is realized.
A tunable self-seeded multi-wavelength Brillouin-erbium fiber laser (BEFL) is proposed and demonstrated based on a few-mode fiber filter (FMFF) with varying temperature. The FMFF configuration is a section of uncoated few-mode fiber (FMF) sandwiched between two up-tapers. As the temperature varies from 25 °C to 125 °C, the transmission spectrum of FMFF moves towards the longer wavelength. The self-excited Brillouin pump is internally achieved by cascaded stimulated Brillouin scattering (SBS) in the single mode fiber (SMF). Then employing the FMFF temperature variation characteristics in the ring cavity fiber laser, the multi-wavelength of the output laser can be tuned, and the tunable range is about 8.0 nm. The generation of up to 15 Brillouin Stokes wavelengths with 16 dB optical signal-to-noise ratio (OSNR) is realized.
2017, 13(6):414-418.
DOI: https://doi.org/10.1007/s11801-017-7182-7
Abstract:
The application of Golay pulse coding technique in spontaneous Brillouin-based distributed temperature sensor based on self-heterodyne detection of Rayleigh and Brillouin scattering is theoretically and experimentally analyzed. The enhancement of system signal to noise ratio (SNR) and reduction of temperature measurement error provided by coding are characterized. By using 16-bit Golay coding, SNR can be improved by about 2.77 dB, and temperature measurement error of the 100 m heated fiber is reduced from 1.4 °C to 0.5 °C with a spatial resolution of 13 m. The results are believed to be beneficial for the performance improvement of self-heterodyne detection Brillouin optical time domain reflectometer.
The application of Golay pulse coding technique in spontaneous Brillouin-based distributed temperature sensor based on self-heterodyne detection of Rayleigh and Brillouin scattering is theoretically and experimentally analyzed. The enhancement of system signal to noise ratio (SNR) and reduction of temperature measurement error provided by coding are characterized. By using 16-bit Golay coding, SNR can be improved by about 2.77 dB, and temperature measurement error of the 100 m heated fiber is reduced from 1.4 °C to 0.5 °C with a spatial resolution of 13 m. The results are believed to be beneficial for the performance improvement of self-heterodyne detection Brillouin optical time domain reflectometer.
2017, 13(6):419-422.
DOI: https://doi.org/10.1007/s11801-017-7183-6
Abstract:
The ultraviolet (UV) band edge photorefractivity of LiNbO3:Zr at 325 nm has been investigated. The experimental results show that the resistance against photorefraction at 325 nm is quite obvious but not as strong as that at 351 nm, when the doping concentration of Zr reaches 2.0 mol%. It is reported that the photorefractivity in other tetravalently doped LiNbO3 crystals, such as LiNbO3:Hf and LiNbO3:Sn, is enhanced dramatically with doping concentration over threshold. Here we give an explicit explanation on such seemly conflicting behaviors of tetravalently doped LiNbO3, which is ascribed to the combined effect of increased photoconductivity and the absorption strength of the band edge photorefractive centers.
The ultraviolet (UV) band edge photorefractivity of LiNbO3:Zr at 325 nm has been investigated. The experimental results show that the resistance against photorefraction at 325 nm is quite obvious but not as strong as that at 351 nm, when the doping concentration of Zr reaches 2.0 mol%. It is reported that the photorefractivity in other tetravalently doped LiNbO3 crystals, such as LiNbO3:Hf and LiNbO3:Sn, is enhanced dramatically with doping concentration over threshold. Here we give an explicit explanation on such seemly conflicting behaviors of tetravalently doped LiNbO3, which is ascribed to the combined effect of increased photoconductivity and the absorption strength of the band edge photorefractive centers.
2017, 13(6):423-426.
DOI: https://doi.org/10.1007/s11801-017-7159-6
Abstract:
A 10 Gbit/s 16-km-long reconfigurable wavelength-division-multiplexing passive optical network (WDM-PON) is presented empowered by a low-cost multi-channel directly modulated laser (DML) module. Compared with the case using discrete devices in conventional scheme, the proposed DML module provides a cost-effective solution with reduced complexity. The clear eye diagram and the bit error rate (BER) of less than 2×10-7 with a sensitivity of −7 dBm are obtained. Due to the special packaging design, the crosstalk between channels under condition of simultaneous operation can be negligible.
A 10 Gbit/s 16-km-long reconfigurable wavelength-division-multiplexing passive optical network (WDM-PON) is presented empowered by a low-cost multi-channel directly modulated laser (DML) module. Compared with the case using discrete devices in conventional scheme, the proposed DML module provides a cost-effective solution with reduced complexity. The clear eye diagram and the bit error rate (BER) of less than 2×10-7 with a sensitivity of −7 dBm are obtained. Due to the special packaging design, the crosstalk between channels under condition of simultaneous operation can be negligible.
2017, 13(6):427-431.
DOI: https://doi.org/10.1007/s11801-017-7195-2
Abstract:
Time synchronization techniques, especially on the pulse per second (PPS) temporal basis, have attracted growing research interests in recent years. In this paper, we have proposed and experimentally demonstrated a high-precision two-way time transfer (TWTT) system to realize long-distance dissemination of 1 PPS signal generated by a hydrogen maser. A dense-wavelength-division-multiplexing (DWDM) system and bi-directional erbium-doped fiber amplifiers (Bi-EDFAs) have also been adopted to suppress the impact of Rayleigh backscattering and optimize the signal to noise ratio (SNR) as well. We have theoretically analyzed the systematic delay in detail. The ultimate root mean square (RMS) variation of time synchronization accuracy is sub-26 ps and the time deviation can be reduced to as low as 1.2 ps at 100 s and 0.253 ps at 12 000 s, respectively.
Time synchronization techniques, especially on the pulse per second (PPS) temporal basis, have attracted growing research interests in recent years. In this paper, we have proposed and experimentally demonstrated a high-precision two-way time transfer (TWTT) system to realize long-distance dissemination of 1 PPS signal generated by a hydrogen maser. A dense-wavelength-division-multiplexing (DWDM) system and bi-directional erbium-doped fiber amplifiers (Bi-EDFAs) have also been adopted to suppress the impact of Rayleigh backscattering and optimize the signal to noise ratio (SNR) as well. We have theoretically analyzed the systematic delay in detail. The ultimate root mean square (RMS) variation of time synchronization accuracy is sub-26 ps and the time deviation can be reduced to as low as 1.2 ps at 100 s and 0.253 ps at 12 000 s, respectively.
2017, 13(6):432-435.
DOI: https://doi.org/10.1007/s11801-017-7140-4
Abstract:
A new noise reduction method based on ensemble empirical mode decomposition (EEMD) is proposed to improve the detection effect for fluorescence spectra. Polycyclic aromatic hydrocarbons (PAHs) pollutants, as a kind of important current environmental pollution source, are highly oncogenic. Using the fluorescence spectroscopy method, the PAHs pollutants can be detected. However, instrument will produce noise in the experiment. Weak fluorescent signals can be affected by noise, so we propose a way to denoise and improve the detection effect. Firstly, we use fluorescence spectrometer to detect PAHs to obtain fluorescence spectra. Subsequently, noises are reduced by EEMD algorithm. Finally, the experiment results show the proposed method is feasible.
A new noise reduction method based on ensemble empirical mode decomposition (EEMD) is proposed to improve the detection effect for fluorescence spectra. Polycyclic aromatic hydrocarbons (PAHs) pollutants, as a kind of important current environmental pollution source, are highly oncogenic. Using the fluorescence spectroscopy method, the PAHs pollutants can be detected. However, instrument will produce noise in the experiment. Weak fluorescent signals can be affected by noise, so we propose a way to denoise and improve the detection effect. Firstly, we use fluorescence spectrometer to detect PAHs to obtain fluorescence spectra. Subsequently, noises are reduced by EEMD algorithm. Finally, the experiment results show the proposed method is feasible.
2017, 13(6):436-438.
DOI: https://doi.org/10.1007/s11801-017-7145-z
Abstract:
A nonlinearity measurement of the charge-coupled device (CCD) array spectrometer using flux addition and comparison method is described. The light with various colors from the colorful light emitting diode (LED) light source is applied to measure the nonlinearity of the spectrometer at different wavelengths, respectively. An high-end CCD array spectrometer is tested. For colorful LED light sources, the nonlinearity factors of the CCD array spectrometer (absolute value) are as follows:k<0.8% for white light, k <1.1% for red light, k <2.2% for green light and k<4.7% for blue light. By using those quasi-monochromatic light sources, it is shown that the nonlinearity depends on the wavelength. It is important to be wariness about the spectral nonlinearity and related uncertainty evaluation when the narrow-band light source is tested.
A nonlinearity measurement of the charge-coupled device (CCD) array spectrometer using flux addition and comparison method is described. The light with various colors from the colorful light emitting diode (LED) light source is applied to measure the nonlinearity of the spectrometer at different wavelengths, respectively. An high-end CCD array spectrometer is tested. For colorful LED light sources, the nonlinearity factors of the CCD array spectrometer (absolute value) are as follows:k<0.8% for white light, k <1.1% for red light, k <2.2% for green light and k<4.7% for blue light. By using those quasi-monochromatic light sources, it is shown that the nonlinearity depends on the wavelength. It is important to be wariness about the spectral nonlinearity and related uncertainty evaluation when the narrow-band light source is tested.
2017, 13(6):439-443.
DOI: https://doi.org/10.1007/s11801-017-7143-1
Abstract:
To make use of the prior knowledge of the image more effectively and restore more details of the edges and structures, a novel sparse coding objective function is proposed by applying the principle of the non-local similarity and manifold learning on the basis of super-resolution algorithm via sparse representation. Firstly, the non-local similarity regularization term is constructed by using the similar image patches to preserve the edge information. Then, the manifold learning regularization term is constructed by utilizing the locally linear embedding approach to enhance the structural information. The experimental results validate that the proposed algorithm has a significant improvement compared with several super-resolution algorithms in terms of the subjective visual effect and objective evaluation indices.
To make use of the prior knowledge of the image more effectively and restore more details of the edges and structures, a novel sparse coding objective function is proposed by applying the principle of the non-local similarity and manifold learning on the basis of super-resolution algorithm via sparse representation. Firstly, the non-local similarity regularization term is constructed by using the similar image patches to preserve the edge information. Then, the manifold learning regularization term is constructed by utilizing the locally linear embedding approach to enhance the structural information. The experimental results validate that the proposed algorithm has a significant improvement compared with several super-resolution algorithms in terms of the subjective visual effect and objective evaluation indices.
2017, 13(6):444-447.
DOI: https://doi.org/10.1007/s11801-017-7168-5
Abstract:
In order to solve the problem of low recognition rate of traditional feature extraction operators under low-resolution images, a novel algorithm of expression recognition is proposed, named central oblique average center-symmetric local binary pattern (CS-LBP) with adaptive threshold (ATCS-LBP). Firstly, the features of face images can be extracted by the proposed operator after pretreatment. Secondly, the obtained feature image is divided into blocks. Thirdly, the histogram of each block is computed independently and all histograms can be connected serially to create a final feature vector. Finally, expression classification is achieved by using support vector machine (SVM) classifier. Experimental results on Japanese female facial expression (JAFFE) database show that the proposed algorithm can achieve a recognition rate of 81.9% when the resolution is as low as 16×16, which is much better than that of the traditional feature extraction operators.
In order to solve the problem of low recognition rate of traditional feature extraction operators under low-resolution images, a novel algorithm of expression recognition is proposed, named central oblique average center-symmetric local binary pattern (CS-LBP) with adaptive threshold (ATCS-LBP). Firstly, the features of face images can be extracted by the proposed operator after pretreatment. Secondly, the obtained feature image is divided into blocks. Thirdly, the histogram of each block is computed independently and all histograms can be connected serially to create a final feature vector. Finally, expression classification is achieved by using support vector machine (SVM) classifier. Experimental results on Japanese female facial expression (JAFFE) database show that the proposed algorithm can achieve a recognition rate of 81.9% when the resolution is as low as 16×16, which is much better than that of the traditional feature extraction operators.
2017, 13(6):448-451.
DOI: https://doi.org/10.1007/s11801-017-7185-4
Abstract:
In this paper, a novel framework, named as global-local feature attention network with reranking strategy (GLAN-RS), is presented for image captioning task. Rather than only adopting unitary visual information in the classical models, GLAN-RS explores the attention mechanism to capture local convolutional salient image maps. Furthermore, we adopt reranking strategy to adjust the priority of the candidate captions and select the best one. The proposed model is verified using the Microsoft Common Objects in Context (MSCOCO) benchmark dataset across seven standard evaluation metrics. Experimental results show that GLAN-RS significantly outperforms the state-of-the-art approaches, such as multimodal recurrent neural network (MRNN) and Google NIC, which gets an improvement of 20% in terms of BLEU4 score and 13 points in terms of CIDER score.
In this paper, a novel framework, named as global-local feature attention network with reranking strategy (GLAN-RS), is presented for image captioning task. Rather than only adopting unitary visual information in the classical models, GLAN-RS explores the attention mechanism to capture local convolutional salient image maps. Furthermore, we adopt reranking strategy to adjust the priority of the candidate captions and select the best one. The proposed model is verified using the Microsoft Common Objects in Context (MSCOCO) benchmark dataset across seven standard evaluation metrics. Experimental results show that GLAN-RS significantly outperforms the state-of-the-art approaches, such as multimodal recurrent neural network (MRNN) and Google NIC, which gets an improvement of 20% in terms of BLEU4 score and 13 points in terms of CIDER score.
2017, 13(6):452-456.
DOI: https://doi.org/10.1007/s11801-017-7189-0
Abstract:
Due to the lack of enough information to solve the equation of image degradation model, existing defogging methods generally introduce some parameters and set these values fixed. Inappropriate parameter setting leads to difficulty in obtaining the best defogging results for different input foggy images. Therefore, a single image defogging algorithm based on particle swarm optimization (PSO) is proposed in this letter to adaptively and automatically select optimal parameter values for image defogging algorithms. The proposed method is applied to two representative defogging algorithms by selecting the two main parameters and optimizing them using the PSO algorithm. Comparative study and qualitative evaluation demonstrate that the better quality results are obtained by using the proposed parameter selection method.
Due to the lack of enough information to solve the equation of image degradation model, existing defogging methods generally introduce some parameters and set these values fixed. Inappropriate parameter setting leads to difficulty in obtaining the best defogging results for different input foggy images. Therefore, a single image defogging algorithm based on particle swarm optimization (PSO) is proposed in this letter to adaptively and automatically select optimal parameter values for image defogging algorithms. The proposed method is applied to two representative defogging algorithms by selecting the two main parameters and optimizing them using the PSO algorithm. Comparative study and qualitative evaluation demonstrate that the better quality results are obtained by using the proposed parameter selection method.
2017, 13(6):457-461.
DOI: https://doi.org/10.1007/s11801-017-7197-0
Abstract:
Text extraction is an important initial step in digitizing the historical documents. In this paper, we present a text extraction method for historical Tibetan document images based on block projections. The task of text extraction is considered as text area detection and location problem. The images are divided equally into blocks and the blocks are filtered by the information of the categories of connected components and corner point density. By analyzing the filtered blocks’ projections, the approximate text areas can be located, and the text regions are extracted. Experiments on the dataset of historical Tibetan documents demonstrate the effectiveness of the proposed method.
Text extraction is an important initial step in digitizing the historical documents. In this paper, we present a text extraction method for historical Tibetan document images based on block projections. The task of text extraction is considered as text area detection and location problem. The images are divided equally into blocks and the blocks are filtered by the information of the categories of connected components and corner point density. By analyzing the filtered blocks’ projections, the approximate text areas can be located, and the text regions are extracted. Experiments on the dataset of historical Tibetan documents demonstrate the effectiveness of the proposed method.
2017, 13(6):462-465.
DOI: https://doi.org/10.1007/s11801-017-7198-z
Abstract:
Traditional hand-crafted features for representing local image patches are evolving into current data-driven and learning-based image feature, but learning a robust and discriminative descriptor which is capable of controlling various patch-level computer vision tasks is still an open problem. In this work, we propose a novel deep convolutional neural network (CNN) to learn local feature descriptors. We utilize the quadruplets with positive and negative training samples, together with a constraint to restrict the intra-class variance, to learn good discriminative CNN representations. Compared with previous works, our model reduces the overlap in feature space between corresponding and non-corresponding patch pairs, and mitigates margin varying problem caused by commonly used triplet loss. We demonstrate that our method achieves better embedding result than some latest works, like PN-Net and TN-TG, on benchmark dataset.
Traditional hand-crafted features for representing local image patches are evolving into current data-driven and learning-based image feature, but learning a robust and discriminative descriptor which is capable of controlling various patch-level computer vision tasks is still an open problem. In this work, we propose a novel deep convolutional neural network (CNN) to learn local feature descriptors. We utilize the quadruplets with positive and negative training samples, together with a constraint to restrict the intra-class variance, to learn good discriminative CNN representations. Compared with previous works, our model reduces the overlap in feature space between corresponding and non-corresponding patch pairs, and mitigates margin varying problem caused by commonly used triplet loss. We demonstrate that our method achieves better embedding result than some latest works, like PN-Net and TN-TG, on benchmark dataset.
2017, 13(6):466-470.
DOI: https://doi.org/10.1007/s11801-017-7201-8
Abstract:
High-quality insulator region proposals play important roles in the process of transmission line inspection images. A generation method of insulator region proposals based on edge boxes is proposed in this paper, and edge boxes are applied to the localization of insulators in inspection images creatively. We take a series of operations to generate insulator region proposals:K-means cluster is used on curvature scale space (CSS) points extracted from edge images, the most appropriate cluster number is chosen, and the circle is drawn on the insulator subclass. We consider the characteristics of insulators’ edge images, and combine these characteristics with edge boxes. As a result, more insulator region proposals are displayed. The experimental results show that our method can effectively reduce the interference area, meanwhile, has high quality of region proposals with fast calculation speed.
High-quality insulator region proposals play important roles in the process of transmission line inspection images. A generation method of insulator region proposals based on edge boxes is proposed in this paper, and edge boxes are applied to the localization of insulators in inspection images creatively. We take a series of operations to generate insulator region proposals:K-means cluster is used on curvature scale space (CSS) points extracted from edge images, the most appropriate cluster number is chosen, and the circle is drawn on the insulator subclass. We consider the characteristics of insulators’ edge images, and combine these characteristics with edge boxes. As a result, more insulator region proposals are displayed. The experimental results show that our method can effectively reduce the interference area, meanwhile, has high quality of region proposals with fast calculation speed.
2017, 13(6):471-475.
DOI: https://doi.org/10.1007/s11801-017-7203-6
Abstract:
A way of embedded learning convolution neural network (ELCNN) based on the image content is proposed to evaluate the image aesthetic quality in this paper. Our approach can not only solve the problem of small-scale data but also score the image aesthetic quality. First, we chose Alexnet and VGG_S to compare for confirming which is more suitable for this image aesthetic quality evaluation task. Second, to further boost the image aesthetic quality classification performance, we employ the image content to train aesthetic quality classification models. But the training samples become smaller and only using once fine-tuning cannot make full use of the small-scale data set. Third, to solve the problem in second step, a way of using twice fine-tuning continually based on the aesthetic quality label and content label respective is proposed, the classification probability of the trained CNN models is used to evaluate the image aesthetic quality. The experiments are carried on the small-scale data set of Photo Quality. The experiment results show that the classification accuracy rates of our approach are higher than the existing image aesthetic quality evaluation approaches.
A way of embedded learning convolution neural network (ELCNN) based on the image content is proposed to evaluate the image aesthetic quality in this paper. Our approach can not only solve the problem of small-scale data but also score the image aesthetic quality. First, we chose Alexnet and VGG_S to compare for confirming which is more suitable for this image aesthetic quality evaluation task. Second, to further boost the image aesthetic quality classification performance, we employ the image content to train aesthetic quality classification models. But the training samples become smaller and only using once fine-tuning cannot make full use of the small-scale data set. Third, to solve the problem in second step, a way of using twice fine-tuning continually based on the aesthetic quality label and content label respective is proposed, the classification probability of the trained CNN models is used to evaluate the image aesthetic quality. The experiments are carried on the small-scale data set of Photo Quality. The experiment results show that the classification accuracy rates of our approach are higher than the existing image aesthetic quality evaluation approaches.
2017, 13(6):476-480.
DOI: https://doi.org/10.1007/s11801-017-7209-0
Abstract:
Traffic sign recognition (TSR) is an important component of automated driving systems. It is a rather challenging task to design a high-performance classifier for the TSR system. In this paper, we propose a new method for TSR system based on deep convolutional neural network. In order to enhance the expression of the network, a novel structure (dubbed block-layer below) which combines network-in-network and residual connection is designed. Our network has 10 layers with parameters (block-layer seen as a single layer): the first seven are alternate convolutional layers and block-layers, and the remaining three are fully-connected layers. We train our TSR network on the German traffic sign recognition benchmark (GTSRB) dataset. To reduce overfitting, we perform data augmentation on the training images and employ a regularization method named “dropout”. The activation function we employ in our network adopts scaled exponential linear units (SELUs), which can induce self-normalizing properties. To speed up the training, we use an efficient GPU to accelerate the convolutional operation. On the test dataset of GTSRB, we achieve the accuracy rate of 99.67%, exceed-ing the state-of-the-art results.
Traffic sign recognition (TSR) is an important component of automated driving systems. It is a rather challenging task to design a high-performance classifier for the TSR system. In this paper, we propose a new method for TSR system based on deep convolutional neural network. In order to enhance the expression of the network, a novel structure (dubbed block-layer below) which combines network-in-network and residual connection is designed. Our network has 10 layers with parameters (block-layer seen as a single layer): the first seven are alternate convolutional layers and block-layers, and the remaining three are fully-connected layers. We train our TSR network on the German traffic sign recognition benchmark (GTSRB) dataset. To reduce overfitting, we perform data augmentation on the training images and employ a regularization method named “dropout”. The activation function we employ in our network adopts scaled exponential linear units (SELUs), which can induce self-normalizing properties. To speed up the training, we use an efficient GPU to accelerate the convolutional operation. On the test dataset of GTSRB, we achieve the accuracy rate of 99.67%, exceed-ing the state-of-the-art results.
External Links