Research On Product Testing Technology Based On Compressive Sensing

With the improvement of production technology and the requirements for productquality, People must pay close attention to the accuracy of protect testing. For example,there are key products with enormous data and complex assembled structure in the field ofaerospace, national defense and high-speed rail. Due to the complexity of the production,collected enormous data bring difficulty to online real-time detection, which can’t meet theneed of production. In response to these problems, this article mainly uses the theory ofcompressed sensing, spare representation and spare decomposition to finish classificationrecognition and defects detection of products.First of all, the sparse representation algorithm of classification and recognition based oncompressing Sensing is to divide the collected sequence images after pretreatment into threecategories, then extracting some randomly from three sample images as training samples andthe remaining as test samples. By sparse representation on the test samples, L-1norm, dualaffine method, Solve Homotopy and OMP are used to obtain the most sparse solution, whichhas obvious category information, so it is easy to categorize product images and judge thetype of products.Secondly, after the classification of products, it needs to detect whether the product isdefective. The paper adopts sparse decomposition algorithm of comprehensive sensing todecompose product images to get the background image and detect image by using differentbackground dictionary and detect dictionary. Then it can be determined whether a productqualified according to the characteristic of the detect image.Experimental results show that using sparse representation has high recognition rate forclassification of product. In order to increase the recognition efficiency, it also can classifyproduct through PCA dimension reduction. The sparse decomposition algorithm canseparate its background image of product well, but the detect images are different because ofthe differences in the nature of detects. If the noise is added on the product, it can becompletely separated to achieve the goal of defects detection.