Predicting the suitability of lateritic soil type for low cost sustainable housing with image recognition and machine learning techniques

Abstract

From a sustainability point of view, laterites-compressed earth bricks (LCEB) are a promising substitute for building structures in place of the conventional concrete masonry units. On the other hand, techniques for identifying and classifying laterites soil for compressed earth bricks (CEB) production are still relying on direct human expertise or ‘experts’. Human experts exploit direct visual inspection and other basic senses such as smelling, touching or nibbling to generate a form of binomial classification, i.e. suitable or unsuitable. The source of predictive power is otherwise supposed to be found in color, scent, texture or combinations of these. Lack of clarity regarding the actual method and the possible explanatory mechanisms lead to 1) difficulties to train other people into the skills and 2) might also add to apathy to using CEB masonry units for housing. Here we systematize the selection method of experts. We chose imaging analysis techniques based on 1) easiness in image acquisition (Digital Camera) and 2) availability of machine learning and statistical techniques. We find that most of the predictive power of the ‘expert’ can be packed into visual inspection by demonstrating that with image analysis alone we get a 98% match. This makes it practically unnecessary the study of any other ‘expert’ skills and provides a method to alleviate the housing problems dealing with material construction in the developing world.