Enhanced Corrosion Detection By Means of Multispectral Imaging and Postprocessing Techniques

W.V. Giegerich, D.C. Fedorishin, L. Forte III, L. Christie, P.J. Schneider, K.W. Oh
University at Buffalo,
United States

Keywords: multispectral imaging, corrosion detection, post processing, image enhancement


This research focuses on the implementation of an enhanced corrosion detection algorithm, fused with multispectral imaging techniques and post-processing image filters. By pairing a series of light filters with image enhancement software, we were able to increase the performance of a corrosion detection algorithm in both recall and precision, increasing the F1-Score, meaning an increase in accuracy of corroded pixel detection by 82%. Prior Work: Corrosion is the natural process in which metal undergoes oxidation often causing it to degrade over a period of time. Frequently seen in industrial applications suchs as civil structures, machinery equipment, and automotive, it is also prevalent in MEMS type devices and sensor technology [1]. Manually identifying rust requires a trained eye and significant time to properly identify and locate rust in an image. Efforts have been made to automate that process through image processing algorithms [2]. These algorithms primarily focus on the image characteristics of texture and color utilizing different color spaces such as red green blue (RGB), hue saturation intensity (HSI), hue saturation value HSV. Methods: Enhancing image processing was first done through the implementation of a variety of object illumination techniques, along with image post process filters to enhance points of interest within the image. This allowed for the contrast of different wavelengths of light, and thus color, resulting in a higher corrosion detection accuracy. The process flow for the creation of the ground truth image can be seen in Figure 1. Once established, the image went through a combination of post processing enhancements, using “Polarr Photo Editor”, ranging from saturation, sharpness, contrast, color, vibrance, luminance, dehase, clarity, and highlight. In addition, by focusing on varying spectrums of light between 580-660nm, and neglecting spectrums outside that range, key areas of interest were able to be highlighted. The resulting image is then passed through the corrosion detection algorithm where the hue saturation and value color pixel score was calculated (Figure 2.). Experiments & Results: Corroded objects were imaged using an RGB camera and processed with nine varying post image filters, and implemented into the corrosion detection algorithm. The goal being to determine which combination of methods yields the highest level of detection accuracy, with respect to the ground truth image. The algorithm's performance, being its ability to identify corroded pixels, was quantified utilizing an F1 score which is calculated using a combination of both precision and recall measurements. Quantification of the image processing technique’s performance, as a function of the number of identified corrosion pixels, is shown in Table 1. It was demonstrated that by combining four filters at selected scales (vibrance -25, dehase +75, highlight -100, luminance -100) yielded an 82% increase in an F1 score.