The convolutional neural network (CNN) has recently achieved great breakthroughs in many computer vision tasks. However, its application in fabric texture defects classification has not been thoroughly researched. To this end, this paper carries out a research on its application based on the CNN model. Meanwhile, since the CNN cannot achieve good classification accuracy in small sample sizes, a new method combining compressive sensing and the convolutional neural network (CS-CNN) is proposed. Specifically, this paper uses the compressive sampling theorem to compress and augment the data in small sample sizes; then the CNN can be employed to classify the data features directly from compressive sampling; finally, we use the test data to verify the classification performance of the method. The explanatory experimental results demonstrate that, in comparison with the state-of-the-art methods for running time, our CS-CNN approach can effectively improve the classification accuracy in fabric defect samples, even with a small number of defect samples.

No abstract

No abstract

BackgroundCurrently, many imaging photoplethysmography (IPPG) researches have reported non-contact measurements of physiological parameters, such as heart rate (HR), respiratory rate (RR), etc. However, it is accepted that only HR measurement has been mature for applications, and other estimations are relatively incapable for reliable applications. Thus, it is worth keeping on persistent studies. Besides, there are some issues commonly involved in these approaches need to be explored further. For example, motion artifact attenuation, an intractable problem, which is being attempted to be resolved by sophisticated video tracking and detection algorithms.MethodsThis paper proposed a blind source separation-based method that could synchronously measure RR and HR in non-contact way. A dual region of interest on facial video image was selected to yield 6-channels Red/Green/Blue signals. By applying Second-Order Blind Identification algorithm to those signals generated above, we obtained 6-channels outputs that contain blood volume pulse (BVP) and respiratory motion artifact. We defined this motion artifact as respiratory signal (RS). For the automatic selections of the RS and BVP among these outputs, we devised a kurtosis-based identification strategy, which guarantees the dynamic RR and HR monitoring available.ResultsThe experimental results indicated that, the estimation by the proposed method has an impressive performance compared with the measurement of the commercial medical sensors.ConclusionsThe proposed method achieved dynamic measurement of RR and HR, and the extension and revision of it may have the potentials for more physiological signs detection, such as heart rate variability, eye blinking, nose wrinkling, yawn, as well as other muscular movements. Thus, it might provide a promising approach for IPPG-based applications such as emotion computation and fatigue detection, etc.Electronic supplementary materialThe online version of this article (doi:10.1186/s12938-016-0300-0) contains supplementary material, which is available to authorized users.