Many structures in civil engineering are symmetrical. Crack detection is a critical task in the monitoring and inspection of civil engineering structures. This study implements a lightweight neural network based on the YOLOv4 algorithm to detect concrete surface cracks. In the extraction of backbone and the design of neck and head, the symmetry concept is adopted. The model modules are improved to reduce the depth and complexity of the overall network structure. Meanwhile, the separable convolution is used to realize spatial convolution, and the SPP and PANet modules are improved to reduce the model parameters. The convolutional layer and batch normalization layer are merged to improve the model inference speed. In addition, using the focal loss function for reference, the loss function of object detection network is improved to balance the proportion of the cracks and the background samples. To comprehensively evaluate the performance of the improved method, 10,000 images (256 × 256 pixels in size) of cracks on concrete surfaces are collected to build the database. The improved YOLOv4 model achieves an mAP of 94.09% with 8.04 M and 0.64 GMacs. The results show that the improved model is satisfactory in mAP, and the model size and calculation amount are greatly reduced. This performs better in terms of real-time detection on concrete surface cracks.

Registration between remote sensing images has been a research focus in the field of remote sensing image processing. Most of the existing image registration algorithms applied to feature point matching are derived from image feature extraction methods, such as scale-invariant feature transform (SIFT), speed-up robust features (SURF) and Siamese neural network. Such methods encounter difficulties in achieving accurate image registration where there is a large bias in the image features or no significant feature points. Aiming to solve this problem, this paper proposes an algorithm for multi-source image registration based on geographical location information (GLI). By calculating the geographic location information that corresponds to the pixel in the image, the ideal projected pixel position of the corresponding image is obtained using spatial coordinate transformation. Additionally, the corresponding relationship between the two images is calculated by combining multiple sets of registration points. The simulation experiment illustrates that, under selected common simulation parameters, the average value of the relative registration-point error between the two images is 12.64 pixels, and the registration accuracy of the corresponding ground registration point is higher than 6.5 m. In the registration experiment involving remote sensing images from different sources, the average registration pixel error of this algorithm is 20.92 pixels, and the registration error of the image center is 21.24 pixels. In comparison, the image center registration error given by the convolutional neural network (CNN) is 142.35 pixels after the registration error is manually eliminated. For the registration of homologous and featureless remote sensing images, the SIFT algorithm can only offer one set of registration points for the correct region, and the neural network cannot achieve accurate registration results. The registration accuracy of the presented algorithm is 7.2 pixels, corresponding to a ground registration accuracy of 4.32 m and achieving more accurate registration between featureless images.

Raman spectroscopy has been widely applied in numerous fields including bioanalysis, disease diagnosis, and molecular recognition, owing to its unique advantages of being non-invasive, rapid, and highly specific. However, the acquisition of Raman spectral data is often susceptible to various noise interferences, such as shot noise from the internal detector and dark current noise from the instrument. As a result, the weak Raman signals typically become arduous to discern, which affects the localization and identification of characteristic spectral peaks. This study investigates variational mode decomposition (VMD), empirical wavelet transform (EWT), and integrated encoder-bidirectional long short-term memory (EBiLSTM) modules to propose a neural network algorithm for adaptive denoising of Raman spectra. By combining VMD and EWT, the Raman spectra are decomposed into several sub-sequences, which solves the problem of mode mixing between high-frequency signals and noise in empirical mode decomposition, and significantly reduces the complexity of the original Raman spectral lines. The correlation coefficient between each modal component and the original signal is calculated along with the zero-crossing rate index to categorize the noise and signal sequences. Leveraging the linear differences between the ideal spectral lines and the noisy spectral curves, an encoder-bidirectional long short-term memory (EBiLSTM) denoising network is constructed for hierarchical denoising to extract valid spectral feature information from the high-frequency components, realizing refined adaptive denoising of the Raman spectra. Industry standard objective evaluation metrics on the signal-to-noise ratio and root mean square error are utilized to conduct simulation experiments comparing state-of-the-art algorithms, including empirical mode decomposition, VMD, sliding window averaging, and wavelet thresholding. The experimental results demonstrate that the Raman spectral denoising algorithm combining variational mode decomposition and neural networks improves the denoising performance by 13.38% to 72%, exhibiting higher accuracy and reliability.

This paper presents a multi-scale feature-enhanced water detector using truncated Gaussian clutter (TGCFeWD) for SAR imagery. It aims to improve detection accuracy through adaptive elimination of high-intensity outliers and shadows. High-intensity outliers cause over-estimation of parameters used for water data statistical modeling, resulting in inaccurate thresholds. Furthermore, shadows in SAR imagery have similar digital numbers and statistical parameters to water, making them difficult to distinguish. To address these two key issues, this paper proposes a comprehensive approach comprised of three main components: The Part A is to calculate Gaussian distance based on which the land surfaces or artificial objects can be removed easily by Otsu. The Part B is to expand the difference between water and shadows through feature enhancing. The Part C is to segment water and shadows according to the expanding differences. The proposed TGCFeWD method effectively detects water bodies including seas, rivers and lakes. Compared to several existing methods, TGCFeWD greatly improves water detection accuracy in complex environments. Based on metrics of Ac, F1 and mIoU, TGCFeWD achieves the best performance (92.4%, 82.4%, 80.1% for all data with five water body types) compared to several traditional methods, and even outperforms some neural network-based methods in certain scenarios. The results are validated on the HISEA Flooding dataset.

Signal-to-Noise Ratio (SNR) is the benchmark to evaluate the quality of optical remote sensors. For SNR estimation, most of the traditional methods have complicated processes, low efficiency, and general accuracy. In particular, they are not suitable for the distributed computation on intelligent satellites. Therefore, an intelligent SNR estimation algorithm with strong computing power and more accuracy is urgently needed. Considering the simplicity of distributed deployment and the lightweight goal, our first proposition is to design a convolutional neural network (CNN) similar to VGG (proposed by Visual Geometry Group) to estimate SNR for optical remote sensors. In addition, considering the advantages of multi-branch structures, the second proposition is to train the CNN in a novel method of multi-branch training and parameter-reconstructed inference. In this study, simulated and real remote sensing images with different ground features are utilized to validate the effectiveness of our model and the novel training method. The experimental results show that the novel training method enhances the fitting ability of the network, and the proposed CNN trained in this method has high accuracy and reliable SNR estimation, which achieves a 3.9% RMSE for noise-level-known simulated images. When compared to the accuracy of the reference methods, such as the traditional and typical SNR methods and the denoising convolutional neural network (DnCNN), the performance of the proposed CNN trained in a novel method is the best, which achieves a relative error of 5.5% for hyperspectral images. The study is fit for optical remote sensing images with complicated ground surfaces and different noise levels captured by different optical remote sensors.