The weld defects inspection from radiography films is critical for assuring the serviceability and safety of weld joints. The various limitations of human interpretation made the development of innovative computer-aided techniques for automatic detection from radiography images an interest point of recent studies. The studies of automatic defect inspection are synthetically concluded from three aspects: pre-processing, defect segmentation and defect classification. The achievement and limitations of traditional defect classification method based on the feature extraction, selection and classifier are summarized. Then the applications of novel models based on learning(especially deep learning) were introduced. Finally, the achievement of automation methods were discussed and the challenges of current technology are presented for future research for both weld quality management and computer science researchers.

The development of image sensor and optics enables the application of vision-based techniques to the non-contact dynamic vibration analysis of large-scale structures. As an emerging technology, a vision-based approach allows for remote measuring and does not bring any additional mass to the measuring object compared with traditional contact measurements. In this study, a high-speed vision-based sensor system is developed to extract structure vibration signals in real time. A fast motion extraction algorithm is required for this system because the maximum sampling frequency of the charge-coupled device (CCD) sensor can reach up to 1000 Hz. Two efficient subpixel level motion extraction algorithms, namely the modified Taylor approximation refinement algorithm and the localization refinement algorithm, are integrated into the proposed vision sensor. Quantitative analysis shows that both of the two modified algorithms are at least five times faster than conventional upsampled cross-correlation approaches and achieve satisfactory error performance. The practicability of the developed sensor is evaluated by an experiment in a laboratory environment and a field test. Experimental results indicate that the developed high-speed vision-based sensor system can extract accurate dynamic structure vibration signals by tracking either artificial targets or natural features.

Fringe projection profilometry has been widely used for 3-D measurement, but the gamma nonlinearity severely affects the accuracy. Ideally, the histogram of the undistorted phase is uniform, but that of the distorted phase becomes nonuniform. In practice, adequate sampling pixels are required to calculate the phase histogram from modulated fringe patterns. Therefore, this article proposes an effective shifted-phase histogram equalization (SHE) method for nonlinear correction. Based on the finding that the distorted histogram contains three symmetric valleys in theory, if we shift the wrapped phase by $\pm 2\pi $ /3, the shifted phases will share the same histogram as the original phase. Without additional patterns, we can easily obtain two shifted phases. Combining the original and two shifted phases, the number of sampling pixels will increase by three times, and we can obtain more accurate histogram than the original phase. To make the histogram of the distorted phase be uniform as that of the undistorted phase, we directly apply histogram equalization on the distorted phase to estimate the undistorted phase. Moreover, this article proposes a spline interpolation optimization (SIO) method to eliminate the discretization error caused by histogram equalization. Simulations and experiments have been carried out, and their results indicate that the proposed SHE method performs better than the conventional phase histogram equalization (PHE) method, and the proposed SIO method can further reduce the discretization error.

This note reports an efficient singular value decomposition (SVD)-based vibration extraction approach that recovers sound information in silent high-speed video. A high-speed camera of which frame rates are in the range of 2 kHz–10 kHz is applied to film the vibrating objects. Sub-images cut from video frames are transformed into column vectors and then reconstructed to a new matrix. The SVD of the new matrix produces orthonormal image bases (OIBs) and image projections onto specific OIB can be recovered as understandable acoustical signals. Standard frequencies of 256 Hz and 512 Hz tuning forks are extracted offline from their vibrating surfaces and a 3.35 s speech signal is recovered online from a piece of paper that is stimulated by sound waves within 1 min.

Accurate and rapid recognition of fruit is the guarantee of intelligent persimmon picking. Given the changes in the light and occlusion conditions in a natural environment, this study developed a detection method based on the improved YOLOv5 model. This approach has several critical steps, including optimizing the loss function based on the traditional YOLOv5, combining the centralized feature pyramid (CFP), integrating the convolutional block attention module (CBAM), and adding a small target detection layer. Images of ripe and unripe persimmons were collected from fruit trees. These images were preprocessed to enhance the contrast, and they were then extended by means of image enhancement to increase the robustness of the network. To test the proposed method, several experiments, including detection and comparative experiments, were conducted. From the detection experiments, persimmons in a natural environment could be detected successfully using the proposed model, with the accuracy rate reaching 92.69%, the recall rate reaching 94.05%, and the average accuracy rate reaching 95.53%. Furthermore, from the comparison experiments, the proposed model performed better than the traditional YOLOv5 and single-shot multibox detector (SSD) models, improving the detection accuracy while reducing the leak detection and false detection rate. These findings provide some references for the automatic picking of persimmons.

Detecting defects from weld radiography images is an important topic in the field of nondestructive testing. Many intelligent detection systems are constructed based on computer. Feature extraction is critical for constructing such system to recognize and classify the weld defects. Deep neural networks have powerful ability to learn representative features that are more sensitive to classification. However, a large number of samples are usually required. In this paper, a stacked autoencoder network is used to pretrain a deep neural network with a small dataset. We can learn the hierarchical feature from the network. In addition, two kinds of traditional manual features are extracted from the same set. These features are combined into new fusion feature vectors for classifying different weld defects. Two evaluation methods are used to test the classification performance of these features through several experiments. The results show that deep feature based on stacked autoencoder network performs better than the other features. The classification performance of fusion features is better than single feature.