STATISTICAL ADAPTIVE AND NEURAL NET CLASSIFIERS APPLIED TO SOLID WASTE PROCESSING: A CRITICAL COMPARISON Statistical adaptive and neural network classifiers applied to solid waste processing: a critical comparison

The adoption, in these last years, of specialised equipment or complex processing architectures developed to separate solid waste materials, resulting from the selective collection of solid urban waste (equipment or manufactured goods dismantling at the end of their life cycle) requires more and more control systems able to “qualify” the products during the processing. Such a goal, when implemented “on-line”, is usually realised in two steps. The attributes (physical, chemical, morphological, morphometrical, textural, etc.) of the materials resulting from processing are detected and numerically modelled. The resulting feature vector is then “handled” by a software architecture performing the required recognition/classification procedure and defining the quality of the investigated products. “Feed-back” or “feed-forward” control strategies can be then applied to improve equipment or processing architectures performances. In this paper are analysed and described the problems encountered and the results achieved when statistical adaptive classifiers or a neural net based architectures are adopted to define an “artificial intelligence software unit” able to perform the recognition of several solid waste materials, at industrial recycling processing plant level, starting from their preliminary optical recognition.