This PDF file contains the front matter associated with SPIE Proceedings Volume 12281, including the Title Page, Copyright information, Table of Contents, and Conference Committee listings.

Sparse sampling of spectral components in Segmented Planar Imaging Detector for Electro-Optical Reconnaissance is an essential limiting factor to the imaging resolution. A dictionary learning method is proposed to improve the imaging quality. The images are segmented into patches, and data are extracted directly from small patches and taken as dictionary elements. By training high-and-low resolution image pairs, a coupled dictionary is obtained. The TV/L1 minimization and alternating direction multiplier method are used to restore high-resolution images. In this way, the quality metric RMSE of images is improved from 20.99 to 14.99, and PSNR from 21.69 dB to 24.62 dB.

In the field of compressive sensing spectral imaging, an adaptive coding method based on a-prior knowledge is a way to obtain high-precision scene information. In this paper, we propose a method that uses low-resolution spatial-spectral information to split into homogeneous regions before generating adaptive coding matrices, in response to the shortcomings of most existing adaptive coding methods that use only spatial a priori information to generate coding matrices. The method uses coding devices in a compressive spectral imaging system to obtain spectral a-priori information with low spatial resolution. Based on this a-priori information, an adaptive segmentation method with region merging is used to obtain segmented images with certain regional homogeneity. The adaptive coding theory and this segmentation result are combined to generate the adaptive coding matrix, and then the compressive observation information of the scene and its complementary observation information are obtained. Based on these observations, the scene information with a high spatial resolution is calculated by the reconstruction algorithm. Simulation experiments show that the adaptive compressive coding method based on spectral image region segmentation has advantages in peak signal-to-noise ratio and structural consistency rating indexes compared with traditional adaptive coding methods.

Ethylene is a gaseous hormone involved in the ripening process of various plants. Monitoring ethylene concentration is helpful to understand the ripening situation of plants, which is conducive to the harvesting, storage and transportation of fruits and other crops. Therefore, it is of great significance to apply the detection technology of ethylene concentration to the monitoring of plant ripening process. In this paper, the feasibility of applying photoacoustic spectrum gas detection technology to the monitoring of plant maturation process is explored, and an experimental device for ethylene concentration detection is designed. The modulated infrared light source and non resonant photoacoustic cell are used to produce photoacoustic effect on the ethylene in the mixed gas to be measured, and the weak signal collected by the microphone is extracted by phase-locked amplification technology. By using this device, ppm level of the detection of ethylene concentration in the mixed gas to be measured can be realized.

Patch-based inpainting is widely used in interactive photo editing scenarios. It iteratively fills the target region by searching the candidate from the source region. However, the high computational cost is a long-lasting concern that prevents users from seeing instant results, bringing limitation to its further application. In this paper, we present a novel instant inpainting technique based on a multi-scale framework, pushing the speed to an unprecedented interactive level (around seconds). Our key insights are that filling process becomes very efficient at low scale. Also, scale changes do not significantly affect the match correspondence, allowing the source-target match correspondence to be quickly collected at low scale, and to be delivered as a priori to higher scales. At high scale, thanks to the built-in coherence in natural images, we propose a mechanism named propagation optimization to further fine-tune the match correspondence based on prior, eliminating the side effects caused by scale recovery. Experiments demonstrate that our method is 10-300 times faster while keeping the same image quality as previous works did. We believe our work contributes a new strategy to real-time application of image inpainting techniques.

Polarization imaging technology has a wide range of applications. However, in practice, polarmetric images are greatly affected by noise, which leads to the loss of polarization information. In this paper, we propose, an attention-based method for polarimetric image denoising. This method innovative utilizes the attention based residual dense network and take full advantage of the importance between feature maps. Experimental results show the proposed method can effectively recover the polarization information from the polarimetric images with low signal noise radio and outperforms other existing methods. Especially for the images of the degree of polarization and the angle of polarization, which are quite sensitive to the noise, the proposed attention-based method can well restore the image detail which is covered by noise.

This paper describes a new RXDMTD algorithm based on RX anomaly detection for moving weak and small targets in multispectral image sequences. The proposed algorithm can effectively suppress background clutter and at the same time enhance the moving weak and small targets in multispectral and out-of-time image sequences. The complex background intensity between the two multispectral images changes significantly, which makes it difficult to suppress the background and difficult to extract the target. In this paper, the image sequence is first arranged and combined, and then the RX algorithm is used to enhance the target and using the target’s movement suppresses the background. Experimental results show that the algorithm proposed in this paper has achieved good detection results.

In recent years, with the increasing demand for multi-source information fusion technology, infrared and visible image fusion technology has been developed rapidly and become an important research direction in the optical information processing field. Combing the advantage of LabVIEW and MATLAB, we proposed a new infrared and visible image fusion system in this paper. An infrared and visible light video image fusion system based on LabVIEW and MATLAB is built. To solve the problem of low infrared image resolution and poor image quality, we proposed an infrared image enhancement algorithm. Experiments result show that the algorithm can enhance the edge features of the infrared image while retaining the internal details of the image. A filtering layered fusion algorithm based on wavelet transform and weighting method is proposed. The algorithm uses anisotropic filtering to decompose the image, calculate each layer's fusion weight, and use the wavelet transform and weighting fusion method to obtain the fused image. Both simulation and actual system experiments prove the effectiveness of the system design and the algorithm.

The sea background video has a wide range of applications in the fields of port maritime traffic management, combating illegal fishing vessels, and maritime rescue. However, the target pixel size in the sea background video is quite small, so increasing the resolution of the target has important practical significance. There are a lot of ripples in the sea background video, which leads to poor video super-resolution effect. We propose a video super-resolution algorithm (CARVSR) in sea background based on the channel attention mechanism. The algorithm adds spatio-temporal 3D learning convolution to the fusion module, which suppresses the interference of ripples on super-resolution reconstruction, and adds channel attention mechanism to the reconstruction module, which enhances the feature expression to reconstruction and improves super-resolution reconstruction quality. Experimental results show that the algorithm effectively improves the superresolution reconstruction effect of sea background video.

The structure of the imaging system of the contact image sensor(CIS) line array camera is different from that of the ordinary camera. The sensors in the camera are generally bound and encapsulated by multiple independent imaging sensors. The problem of camera non-uniformity exists not only in the independent sensor, but also between each group of sensors. Therefore, in order to solve the problem of non-uniformity in the imaging of the line array camera of contact image sensor, especially the problem of non-uniformity mutation near the stitching gap of each imaging unit of the line array camera, an optimized stretch imaging correction method based on the traditional two-point correction method was proposed in this paper. In this method, multiple images with different exposure times are taken as sample sets and the least square method is used to find the best correction black and white template. Experiments show that the relative error offset of the optimized algorithm is smaller than that of the traditional two-point correction method under different exposure time, and the overall image uniformity is improved by about 40% compared with the traditional two-point correction method in the whole response range. Response near stitching gap transition is more uniform, stitching gap near relative to the traditional two-point method NU most about 50% drop in value, and analyzed the optimization calculation of two-point method is easy to convert the FPGA logic implementation. In the high real-time and wide swath image processing architecture of the large area contact linear array camera, the correction effect can be improved only by modifying the black and white templates of the traditional two-point correction method.

Vortex beam with helical wavefront has great applications, such as optical communication, optical manipulation, quantum information processing, etc. due to its infinite topological charge. Under disturbance or misalignment, the energy on vortex beam will be cast onto neighbored vortex beams and we get the OAM spectrum, which is essential to the applications of vortex beams. To measure the OAM spectrum, the measurement and reconstruction of the optical field is usually of great importance but an additional reference light is not easy to achieve. Therefore, this paper proposes a self-interferometry scheme suitable for unknown light fields. By incorporating a diffuser into a Sagnac-like loop, we realize a selfinterferometer. Then a two dimensional parallel grating is used to project the interfered field into a Sudoku squares. Tuning the polarization of each square by waveplate and polaroid, five interference patterns are obtained. By performing calculations on the five interference patterns, we can reconstruct the complex field of the original light. Then, a canonical routine, e.g. spiral harmonic expansion, is performed to get the OAM spectrum. This work provides a new method for measuring unknown light field and its OAM spectrum without extra reference light and has potential applications in the accurate detection and analysis of emitted signals in free-space optical communications.

The long wave infrared imaging systems have been used widely in military and civil fields. With the rapid development of semiconductor technology, new architectures of embedded systems are emerging and developing towards the direction of lower power consumption, higher integration and stronger performance. The traditional hardware circuit design scheme of infrared detector adopts single-FPGA solutions or discrete DSP+FPGA solutions. The above two schemes have disadvantages such as hard algorithm transplantation, hard serial processing and too complex hardware circuit. In this paper, a hardware circuit for a long-wave infrared detector based on Xilinx Zynq with SOPC architecture is developed, which includes a detector driving circuit, a detector temperature control circuit, an analog-to-digital conversion circuit, a signal processing circuit, and a digital signal output circuit. When the system is working, the signals generated by the detector are input into Xilinx Zynq after an analog-to-digital conversion operation. Then, after being cached by DDR3 chip, non-uniformity correction, dynamic compression and image enhancement are completed in Xilinx Zynq. Finally, the infrared video is transmitted to the remote PC through the network for real-time display through UDP protocol of gigabit Ethernet. The software on the PC supports functions such as screenshot, video recording and log query. Experiment as results show that a uncooled long-wave infrared detector with the resolution 640×512 are driven by the proposed scheme, which has advantages of sufficient data bandwidth, easy serial processing and simple process of memory control. The system designed by this scheme has good imaging effect and strong extensibility.

Infrared imaging systems have been widely used in military and civil fields. However, the degradation of imaging quality is constantly affected by stripe noises. The traditional mean filtering, median filtering, Gaussian filtering, Wiener filtering and other algorithms have dependence on different noise images, and the image edge is blurred by denoising. The popular bilateral filter takes a lot of calculation due to a two-dimension way and floating point spatial proximity factor. In this paper, an infrared sequence image denoising method based on multi-frame averaging and improved bilateral filtering is introduced. An improved bilateral filter with an integer spatial proximity factor is designed, which is realized by one dimension filtering in horizontal and vertical directions. First, basal stripe in each frame image is removed by two-point correction, and the improved bilateral filter is used to smooth the noise and protect the edge of the image at the same time. Second, random noise is further removed by averaging multi-frame with a set of 25 images. The experimental results show that the proposed denoising method can effectively remove the noise and better maintain the edge structural information of the image.

X-ray imaging is widely used in airports and transportation for security maintaining. Conventional x-ray images often suffer from noise interference, over sharpening or detail loss, especially in areas where multiple objects overlap each other. To overcome the shortcomings of traditional methods, this article presents a method to reveal the details based on convolutional neural network (CNN). We put forward a well-designed super resolution (SR) network exploiting self guided architecture to fuse multi-scale information. At each scale, we adopt residual feature aggregation strategy for extracting representative details. We also find it is beneficial to establish links between high energy (HE) and low energy (LE) images, thus the restored images show more fine textures and better material resolution. The comparison experiments demonstrate that the proposed network outperforms traditional approaches for restoring details and suppressing noise effectively.

Traditional Retinex model-based image enhancement methods require careful design of constraints and parameters to handle this highly ill-conditioned decomposition. With the advancement of deep learning algorithms, low-light image enhancement has also achieved deep processing. However, image enhancement based on the RGB color space model is prone to color distortions when enhancing images under the influence of the correlation of the three primary colors of RGB. In this report we apply the HSV color space technique to a Retinex-based network model. Simulations and experiments show that using HSV space-improved deep neural networks can effectively avoid the color distortion problem.

Synthetic aperture radar (SAR) target recognition is an important part of SAR image interpretation. It has been widely used in the field of national defense and national economy. At present, the feature extraction based on convolution is a local operation in space and time. The convolution kernel extracts the features in the local region with a certain step size. The global information of the picture can only be obtained by increasing the number of convolution layers. However, this method will not only increase the difficulty of model training but also makes the optimization of the network more difficult, and even leads to over fitting. Therefore, this paper proposes a SAR image target recognition algorithm based on GoogLeNet -NB, which combines accurate and effective GoogLeNet framework and non-local block(NB). By adding NB to GoogLeNet framework, we can capture more context information, enhance the correlation between pixels and regions, and improve the representation ability of the network. In order to verify the effectiveness of NB, NB is added in different positions of GoogLeNet framework for experimental comparison. Finally, MSTAR database is used to verify the algorithm. The experimental results show that the recognition effect of the GoogLeNet -NB algorithm model proposed in this paper is better than the traditional Alexnet algorithm and GoogLeNet algorithm. In the adding NB in different positions of GoogLeNet framework, adding NB in the front position of GoogLeNet can reduce the loss of information in the training process and obtain more global information. Therefore, GoogLeNet-preNB algorithm has certain advantages over GoogLeNet-postNB algorithm.

With all-weather, all-time imaging characteristics of synthetic aperture radar (SAR), SAR image is applied widely. However, because of the SAR imaging mechanism, speckle noise is inevitably present in SAR images. In the translationinvariant second-generation bandelet transform (TIBT) domain, SAR image despeckling algorithm using edge detection and feature clustering (CFCM-TIBT) combines edge detection and fuzzy C-main clustering (FCM) operation is an efficient way to reduce speckle noise. However, the edges will be blurred by the algorithm. In order to improve the edge preservation ability and reduce false edge phenomenon. A new algorithm, named enhanced edge detection for SAR images despeckling in TIBT domain (CRFCM-TIBT), is proposed in this letter. It combines CFCM-TIBT and an improved edge detector, named C-RBED, which consists of Canny edge detector and rate-based edge detector (RBED). CRFCM-TIBT better realizes the extraction and separation of details that benefits from the ability of eliminating false edge pixels of C-RBED. The process of CRFCM-TIBT: C-RBED is first utilized to calculate and compare edge direction map (EDM) and edge strong map (ESM) several times. Then, TIBT and FCM are used to decompose and despeckle the edge-removed image, respectively. Finally, add the removed edges to the reconstructed image. In the two scenarios of Bedfordshire and Horse track, the Equivalent Look Number (ENL) and Edge Preservation Index (ESI) of this algorithm are better than traditional Lee filtering, FCM-TIBT and CFCM-TIBT. The experimental results show from the evaluation indicators and visual effects that the method proposed in this paper significantly improves ESI while ensuring a better ENL.

Underwater imaging plays an important role in underwater scientific research, biological monitoring, underwater rescue, and military defense. However, the scattering and absorption of particles in the scattering medium lead to the distortion of the originally ordered wavefront phase and the image degradation. Polarimetric imaging technology has become an effective method for underwater image recovery based on its unique advantages. However, traditional polarimetric techniques require mechanically rotating the linear polarizer to find the two images that are brightest (the backscattered light is maximally through the polarizer) and darkest (the backscattered light is maximally blocked by the polarizer). Most polarimetric imaging techniques combined with conventional methods require the presence of a background region in the image to estimate the characteristics of backscattered light. There are also some polarimetric imaging methods that do not require a background area in the image, but most of them require prior knowledge and complex algorithms. All these greatly affect the real-time application of polarimetric imaging technology. In this paper, a rapid automatic underwater image recovery method based on polarimetric imaging is proposed. We optimize the traditional polarimetric imaging method based on the scattering imaging model and solve the problem that the traditional method requires an unsupervised image quality evaluation index to correct the intermediate parameters and selecting the background region, which realize the scattering suppression for the underwater image without background region. The experimental results show that our method can process a 320,000-pixel image in only 0.007 seconds. The proposed method can be applied to the current polarimetric imaging instrument system and has a good effect on underwater image recovery, and the recovery results are almost without distortion. It has strong robustness and is also applicable to images without background region, which provide a promising application prospect in the field of underwater imaging.