Research On Defect Inspecting Algorithms For Heat Pads Based On Machine Vision

Thermal battery is an important military power,and widely used in missiles,ships,nuclear weapons and civil aviation,etc.As an important part of the heating system,heat pads provide heat to the thermal battery.Thus,the quality of heat pads directly determines if the thermal battery works.The manufacturing process of heat pads is cosist of powder production,mixing,pressing and defect detection.And the first three have been achieved automation and localization,however,the last one has not yet.The detection of the heat pads is done by hand currently,which seriously constrains the production efficiency and the quality of the heat pads.The heat pads are made of a powder that contains kclo4 and iron in this paper.The defect inspection for heat pads is to make a quick distinction between the qualified and defective product and classify the four kinds of defects(arris,burr,impurity and crack).First,the hardware subsystem of the visual system has been designed,and the camera is insight 7050 for Congex.After installation and debugging,the hardware system is used to obtain the heat-pads-images which is favorable for processing.Secondly,an adaptive binarized-thresholding of anti-noise morphological edge has been proposed in the image preprocessing.On the basis of that,a foreground-extraction-algorithm based on the edge of anti-noise and corrosive morphology has been proposed to elimilate the interference caused by the background of the heat pads.In addition,in order to realize a rapid distinction between the qualified and defective product,an algorithm based on the edge of anti-noise corrosion morphology has been proposed.Thirdly,a new image enhancement algorithm based on Gaussian high-frequency-emphsis filtering and an improved Catte_pm model has been proposed for the shadows,highlights,inclusions and texture in the image of heat pads.Then,an improved Minimum error thresholding method has been proposed for the invalidation of the primary algorithm caused by the background of the image.In addition,the recognition-accuracy of the three methods for classification has been compared: statistical characteristics,PCA-SVM and PCA-NET.What's more,the relationship between the dimension of the features and the recognition-accuracy has been studied.Finally,the software subsystem has been developed,and its function and technical indicators has been verified.Experiments show that the proposed algorithm achieves a rapid classification of qualified and defective products.At the same time,it has a high recognition-accuracy for four classes of defects.