A super-resolution reconstruction method of infrared images based on channel attention and transfer learning was proposed to solve the problems of low resolution and low quality of infrared images. In this method, a deep convolutional neural network is designed to enhance the learning ability of the network by introducing the channel attention mechanism, and the residual learning method is used to mitigate the problem of gradient explosion or disappearance and to accelerate the convergence of the network. Because high-quality infrared images are difficult to collect and insufficient in number, so this method is divided into two steps: the first step is to use natural images to pre-train the neural network model, and the second step is to use transfer learning knowledge to fine-tune the pre-trained model’s parameters with a small number of high-quality infrared images to make the model better in reconstructing the infrared image. Finally, a multi-scale detail boosting filter is added to improve the visual effect of the reconstructed infrared image. Experiments on Set5 and Set14 datasets as well as infrared images show that the deepening network depth and introducing channel attention mechanism can improve the effect of super-resolution reconstruction, transfer learning can well solve the problem of insufficient number of infrared image samples, and multi-scale detail boosting filter can improve the details and increase the amount of information of the reconstruction image.

In order to solve the problems of low work efficiency and safety caused by the inability of human eyes to accurately determine the position of the grab during the loading and unloading of dry bulk cargo by portal crane, a method of grab detection based on deep learning is proposed for the first time. The improved deep convolution neural network (YOLOv3-tiny) is used to train and test on the data set of grab, and then to learn its internal feature representation. The experimental results show that the detection method based on deep learning can achieve a detection speed of 45 frames per second and a recall rate of 95.78%. It can meet the real-time and accuracy of detection, and improve the safety and efficiency of work in the industrial field.

This paper proposes a correlation filter tracking algorithm based on adaptive spatial regularization and aberrance repression aiming at the problem that the spatial regularization weight of the background-aware correlation filter is fixed and does not adapt to the change of the target, and the problem that enlarging search area may introduce background noise, decreasing the discrimination ability of filters. First, FHOG features, CN features, and gray features are extracted to enhance the algorithm's ability to express the target. Second, aberrance repression terms are added to the target function to constrain the response map of the current frame, and to enhance the filter's discrimination ability to alleviate the filter model degradation. Finally, adaptive spatial regularization terms are added to the objective function to make the spatial regularization weights being updated as the objective changes, so that the filter can make full use of the target's diversity information. This paper involves experiments on the public data sets OTB-2013, OTB-2015 and VOT2016 to evaluate the proposed algorithm. The experimental results show that the speed of the algorithm used in this paper is 20 frames/s, evaluation indicators such as distance accuracy and success rate are superior to comparison algorithms, and it has good robustness in a variety of complex scenarios such as occlusion, background interference, and rotation changes.

In the power system, it is difficult to detect the insulator's deterioration in operation. Aiming at this problem, this thesis applies the convolution neural network algorithm to evaluate the insulator's deterioration degree based on the deep analysis of the principle and structure of the convolution neural network model. Firstly, the power frequency flashover test was conducted on the insulator to produce three states as follows: no discharge, weak discharge, and strong discharge. Moreover, the Ultraviolet imager was applied to collect the insulator's ultraviolet images in different discharge state to establish the ultraviolet images sample library. Subsequently, the VGGNet framework neural network algorithm was applied to perform the classification training and the state-prediction evaluation on the samples so as to eventually achieve the purpose of judging whether the insulator is degraded. From the experimental results, it can be seen that the accuracy rate of the algorithm is as high as 98.4%, which has broad application prospects in the insulator's degradation detection. Furthermore, it provides a mentality for the reliability detection of other power equipments.

We propose and experimentally demonstrate a novel in-band optical signal-to-noise ratio (OSNR) monitoring technique that uses a commercially available widely tunable optical bandpass filter to sample the measured optical power as input features of Gaussian process regression (GPR) can accurately estimate the large dynamic range OSNR and is not affected by the configuration of the optical link, and has the characteristics of distributed and low cost. Experimental results for 32 Gbaud PDM-16QAM signals demonstrate OSNR monitoring with the root mean squared error (RMSE) of 0.429 dB and the mean absolute error (MAE) of 0.294 dB within a large OSNR range of -1 dB~30 dB. Moreover, our proposed technique is proved to be insensitive to chromatic dispersion, polarization mode dispersion, nonlinear effect, and cascaded filtering effect (CFE). Furthermore, our proposed technique has the potential to be employed for link monitoring at the intermediation nodes without knowing the transmission information and is more convenient to operate because no calibration is required.

It is far essential to properly recognize the vehicle identification number (VIN) engraved on the car frame for vehicle surveillance and identification. In this paper, we propose an algorithm for recognizing rotational VIN images based on neural network which incorporates two components: VIN detection and VIN recognition. Firstly, with lightweight neural network and text segmentation based on EAST, we attain efficient and excellent VIN detection performance. Secondly, the VIN recognition is regarded as a sequence classification problem. By means of connecting sequential classifiers, we predict VIN characters directly and precisely. For validating our algorithm, we collect a VIN dataset, which contains raw rotational VIN images and horizontal VIN images. Experimental results show that the algorithm we proposed achieves good performance on VIN detection and VIN recognition in real time.

The background of the EL image of a photovoltaic cell under electroluminescence (EL) presents complex non-uniform texture features, and there are grain pseudo-defects similar to cracks. At the same time, the cracks appear as multi-scale features with various shapes. The above mentioned difficulties have presented great challenges for the detection task. Therefore, this paper proposes a multi-scale Faster-RCNN model that integrates attention. On the one hand, an improved feature pyramid network is used to obtain multi-scale advanced semantic feature maps to improve the network's feature expression ability of multi-scale crack defects. On the other hand, an improved attention region proposal network A-RPN is adopted to increase the model's attention to crack defects and suppress the characteristics of complex background and grain pseudo-defects. At the same time, in the RPN network training process, a loss function Focal loss is used to reduce the proportion of simple samples in the training process, so that the model pays more attention to the samples that are difficult to distinguish. Experimental results show that this algorithm improves the accuracy of crack defect detection in EL images, reaching nearly 95%.

Aiming at the low adaptability of blurring noise of target feature points in traditional calibration methods, a calibration method based on the color-coded phase-shifted fringe is proposed. Using a liquid crystal display panel as the calibration target, horizontal and vertical color-coded phase-shifted stripes are displayed in sequence; the orthogonal phase-shifted stripes are obtained by separating color channels; based on the phase-shifteg theory, the intersections of the orthogonal phase truncation lines are calculated as the feature points. After changing the target position multiple times and extracting feature points, the plane-based camera calibration technique is applied to realize the calibration of both the single camera and the binocular system. Furthermore, color-coded phase-shift circles are added to four corners of the target pattern to automatically extract and sort feature points. Accordingly, the effi- ciency of calibration is promoted. The experimental results indicate that when the target image is blurred, the reprojection error of the single-camera calibration is 0.15 pixels, and the standard deviation of the binocular system measurement after calibration is 0.1 mm.

Non-line-of-sight location is an active detection technology which is used to detect the position of objects out of sight by extracting the time of flight. It is a research hotspot in recent years. In order to study the performance differences of mean filter, median filter and Gaussian filter in extracting time of flight, firstly, the energy changing model of photon flight model is optimized by photometry, and then the parameters of the three filtering methods are optimized and analyzed. After that, the adaptability of these three extraction methods to the maximum value judgment method and probability threshold weighted judgment method is analyzed. Finally, the accuracy and stability of these three time extraction algorithms are compared by using the positions of devices and invisible object as variables. The simulation results show that the median filter is suitable for a narrow environment and it has the high accuracy in positioning; the locations with Gaussian filter have good positioning stability and there is a wider selection range of filtering parameters when the signal is processed with Gaussian filter.