Automobile surface defects like scratches or dents occur during the process of manufacturing and cross-border transportation. This will affect consumers’ first impression and the service life of the car itself. In most worldwide automobile industries, the inspection process is mainly performed by human vision, which is unstable and insufficient. The combination of artificial intelligence and the automobile industry shows promise nowadays. However, it is a challenge to inspect such defects in a computer system because of imbalanced illumination, specular highlight reflection, various reflection modes and limited defect features. This paper presents the design and implementation of a novel automatic inspection system (AIS) for automobile surface defects which are the located in or close to style lines, edges and handles. The system consists of image acquisition and image processing devices, operating in a closed environment and noncontact way with four LED light sources. Specifically, we use five plane-array Charge Coupled Device (CCD) cameras to collect images of the five sides of the automobile synchronously. Then the AIS extracts candidate defect regions from the vehicle body image by a multi-scale Hessian matrix fusion method. Finally, candidate defect regions are classified into pseudo-defects, dents and scratches by feature extraction (shape, size, statistics and divergence features) and a support vector machine algorithm. Experimental results demonstrate that automatic inspection system can effectively reduce false detection of pseudo-defects produced by image noise and achieve accuracies of 95.6% in dent defects and 97.1% in scratch defects, which is suitable for customs inspection of imported vehicles.

Deep neural networks (DNNs) have been widely adopted in single image super-resolution (SISR) recently with great success. As a network goes deeper, intermediate features become hierarchical. However, most SISR methods based on DNNs do not make full use of the hierarchical features. The features cannot be read directly by the subsequent layers, therefore, the previous hierarchical information has little influence on the subsequent layer output, and the performance is relatively poor. To address this issue, a novel global dense feature fusion convolutional network (DFFNet) is proposed, which can take full advantage of global intermediate features. Especially, a feature fusion block (FFblock) is introduced as the basic module. Each block can directly read raw global features from previous ones and then learns the feature spatial correlation and channel correlation between features in a holistic way, leading to a continuous global information memory mechanism. Experiments on the benchmark tests show that the proposed method DFFNet achieves favorable performance against the state-of-art methods.