Abstract
The advent of convolutional neural networks (CNNs) has accelerated the progress of computer vision from many aspects. However, the majority of the existing CNNs heavily rely on expensive GPUs (graphics processing units). to support large computations. Therefore, CNNs have not been widely used to inspect surface defects in the manufacturing field yet. In this paper, we develop a compact CNN-based model that not only achieves high performance on tiny defect inspection but can be run on low-frequency CPUs (central processing units). Our model consists of a light-weight (LW) bottleneck and a decoder. By a pyramid of lightweight kernels, the LW bottleneck provides rich features with less computational cost. The decoder is also built in a lightweight way, which consists of an atrous spatial pyramid pooling (ASPP) and depthwise separable convolution layers. These lightweight designs reduce the redundant weights and computation greatly. We train our models on groups of surface datasets. The model can successfully classify/segment surface defects with an Intel i3-4010U CPU within 30 ms. Our model obtains similar accuracy with MobileNetV2 while only has less than its 1/3 FLOPs (floating-point operations per second) and 1/8 weights. Our experiments indicate CNNs can be compact and hardware-friendly for future applications in the automated surface inspection (ASI).
Highlights
For a long time, product quality control relies on manual examination in the manufacturing field
Training deep neural networks on the surface defect datasets may have the issue of data imbalance—first, the image data are difficult to collect, because there only exists a small amount of the products with surface defect; second, the target features is difficult to detect, because the number of surface defect pixels are much less than background pixels
In Reference [58], the segmentation strategy includes two stages: first, an fully convolutional network (FCN) is used to segment defects; second, the 32 × 32 or 64 × 64 patches are cropped around the defect detected by the first stage and a convolutional neural networks (CNNs) is used as a foreground-background classifier to reduce false alarms
Summary
Product quality control relies on manual examination in the manufacturing field. Most proposed defect inspection algorithms for these surfaces are traditional classification or segmentation methods. Reference [12] segments fabric defects by the Sobel and the watershed algorithm These traditional methods involve image preprocessing, feature extraction, feature reduction, and classifier selection, which needs the experience of experts. Traditional algorithms are designed for specific surfaces, effective measures such as recall, precision, and accuracy can barely achieve the standard of industrial applications This long-existing bottleneck has been restraining the growth of ASI. A CNN with high performance needs large datasets to train, while there exist few online datasets for surface defect images To address these limitations, this work introduces our novel lightweight CNN, the compact design remarkably reduces the number of weights and computation while still achieving a high performance. Obtain ∼5% true positive rate improvement on tiny defect detection, by training our model with gradual training and a fine-tuned strategy
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
