Spatiotemporal Image Processing using Line Sensor to Evaluate Fecal Characteristics

Abstract

It is considered to observe fecal characteristics to grasp the state of intestinal environment. Various indicators for evaluating fecal characteristics have been proposed, but it is difficult to evaluate fecal characteristics appropriately due to the influence of the evaluator's subjectivity. Therefore, we propose a method to classify into 6 types objectively using features extracted from line sensor signals while feceses are falling. The captured line sensor signal are combined in the vertical direction in chronological order to generate a spatiotemporal image. From the spatiotemporal image, a total of three features representing the fecal size and ruggedness are extracted to use for the machine learning. As an experimental result for pseudo feceses, the classification accuracy had an average F value = 0.95. As future works, it is necessary to perform evaluation in a real toilet environment and classification for actual feces.