To overcome the problem of fewer sample and uneven distribution of defect type in defect detection of adhesive structure parts, a defect identification approach based on DCGAN and YOLOv5 is proposed. The above problems are solved by fine-tuning the structure and loss function of DCGAN, the generated high-quality defect images and the extended defect dataset are the basis for accurate identification with YOLOv5. The EIOU loss function is utilized in the YOLOv5 network, the mAP value and recall increase by 3.9% and 10.5% compared with the GIOU loss function, but the precision decreases. To solve this problem, the feature extraction capability of the network is enhanced by incorporating the CBAM after the C3 module in the YOLOv5 network. The mAP, precision, and recall of the optimized YOLOv5 algorithm are improved to 78.6%, 77.2%, and 76%, respectively, the precision compared to the original model improved by 10.6%. The results demonstrate that the improved YOLOv5 model can effectively identify defects of adhesive structure.

With the development of manufacturing industry, the high precision size measurement for the holes on the surface of mechanical parts is required. A method for measuring size of the hole based on double projection is proposed in this paper. The CCD camera separately collects grating images which projected onto the part surface in two directions. The phase-shift method is used to calculate the phase of grating images in two directions, and the reconstructed three-dimensional shapes in the two directions are merged to measure the hole dimension. Experiment results show that the measurement accuracy of the thickness is 0.04 mm, and the hole size can reach to 0.1 mm.

In order to meet the increasing demand of non-destructive testing (NDT) in engineering practice, a continuous terahertz NDT platform based on linear scanning has been developed, with a center frequency of 154 GHz and a bandwidth of 56 GHz. This system combines frequency modulation continuous wave (FMCW) radar technology with a continuous scanning structure, as well as a data acquisition platform to provide a non-contact detection method; this is highly efficient and compensates for the shortcomings of traditional methods such as microwave, X-ray, ultrasonic, and others in safety inspection and special detection. In addition, a signal processing method of spectral refinement and correction is proposed in this paper for accurate thickness measurement. The results show that the method has a high accuracy for ABS, PVC, and ceramic matrix composites. By extracting the characteristic parameters, the detection and imaging of prefabricated defects, such as debonding and bubbles in composite materials, have been successfully achieved. This helps to evaluate the internal state of the inspected object more intuitively and further meets the requirements of industrial NDT.

With the increase of glass detection speed, some defects of MapReduce distributed computing framework are exposed, and the processing speed and timeliness cannot meet the requirements of glass-defect detection in industrial technology. Based on the MapReduce distributed computing framework, this paper designs a threshold segmentation method to complete the segmentation of glass-defect images. By improving the replication placement strategy and pipeline scheduling mechanism, the computing and storage are localized, and the timeliness of data processing is accelerated. The experimental results show that the improved MapReduce computing framework has an average increase of 14.8% in processing speed. It can detect the glass ribbon running at 800m/h and also detect the number, position and type of defects on the glass ribbon.

The increasing use of glass fiber reinforced polymer (GFRP) laminates in aircraft creates an urgent demand for nondestructive testing to identify defects such as impact damage and delamination in the materials’ near-surface structure. The High-frequency ultrasonic technique is an effective nondestructive testing technique capable of detecting such defects. However, with the increased delamination depth, it is difficult to accurately characterize the delamination defects in GFRP laminates by traditional characterization methods. And the conventional defect depth detection method takes much time and requires professional discrimination, which is easily affected by the level of inspectors and subjective factors. To address the issue of inaccurate feature extraction and low detection efficiency, we propose a new method for delamination depth recognition in the near-surface of GFRP laminates based on the RQA-MKLSVM model by combining the recurrence quantitative analysis (RQA) feature extraction method with the multi-kernel learning SVM technique to improve the capability of detecting delamination defects on GFRP laminates. Additionally, an image-enhanced method based on the nonlinear transformation function and wavelet multiscale product is used to locate the delaminated areas. Results show that the proposed method achieves accurate recognition of different delamination defects within depths ranging from 0.8 mm to 4.0 mm, and the average recognition rate has reached 95.63%. Compared with the training models based on discrete wavelet transform(DWT) and empirical mode decomposition (EMD), the recognition rate of our method is improved by 6.25% and 4.38%, respectively. And Compared with the two single-core models, the recognition accuracy of our method is improved by 8.55% and 5.00%, respectively.

The problem of image rain removal has drawn widespread attention as the blurry images caused by rain streaks can degrade the performance of many computer vision tasks. Although exiting deep learning-based methods outperform most traditional methods, there are still unresolved issues in terms of performance. In this paper, we propose a novel enhanced attentive generative adversarial network named EAGAN to effectively remove the rain streaks and restore the image structural details at the same time. As rain streaks have different sizes and shapes, EAGAN utilizes a multiscale aggregation attention module (MAAM) to produce an attention map to guide the subsequent network to put conscious attention to rain regions. A symmetrical autoencoder with long-range skip-connections, squeeze-and-excitation (SE) modules, and non-local operation is further utilized to enhance the representation of the network. Finally, spectral normalization and a relativistic generative adversarial network (GAN) are further applied to improve the training stability and deraining performance. Both qualitative and quantitative validations on synthetic and real-world datasets demonstrate that the proposed approach can achieve a competitive performance in comparison with the state-of-the-art methods.