Fuzzy evolutionary shape and shape recognition for log modelling

In a market, which is increasingly challenged by flexible synthetic materials, the sawmill industry is constantly seeking ways to improve its productivity in handling logs of different qualities and sizes. Log modelling and automatic pattern recognition of logs are vital steps in improving the productivity of lumber processing: through the careful analysis of the repeated sawing of a sample log they enable one to obtain information about different lumber production alternatives. However, the current log representation and log recognition models are not adequate for efficient log processing. First, these models are overly static in nature: they deal only with oversimplified external (boundary) log features and rely on static assumptions about log smoothness and log isotropy, and are also based on "static" combinations of Boolean operations on rectangular solids and cylinder-like modelling primitives. Models that rely on strong (static) assumptions about smoothness, isotropy, and combinations of Boolean operations on the above primitives have proved to be fragile, seriously error-prone, and capable of only overly abstracted descriptions of external log shapes. Second, none of these models integrates the non-uniformity of the log interior region (e.g., different patterns of wood types and moisture content) with the fuzzy log boundary features within a single log as well as across logs. Third, using the above modelling primitives for automatic recognition of arbitrary log shapes has not been successful. Since the log's boundary and interior features vary both between and within various tree species, to process (recognize) a new log, the log representation and recognition models must be sufficiently robust and flexible. The limited capabilities of present models lead to a high percentage of fiber loss during the log processing, a high lumber degradation during the drying process, as well as a loss of structural integrity during its subsequent use.; This thesis proposes a new conceptual framework for modelling log shape and its automated recognition, which integrates interior region and boundary features of logs. Guided by the concepts of fuzzy subset entropy and that of the genetic operators, a shape is thought of as being formed by particles interacting via inhibitory and excitatory connections that are subject to an evolution. Based on the concepts of fuzzy subset entropy, the genetic operators, as well as the self-organizing neural network algorithm, the shape is modelled as an adaptive cellular structure--a more realistic shape representation. The cellular structure formation is guided by a range of excitatory and inhibitory interactions among the cells.; The included illustration suggests that the proposed approach can automatically map the log image onto an incrementally growing cellular structure in a manner that can effectively be used in the log recognition process for arbitrary log shapes. The outlined approach makes a conceptual contribution towards the current automated log recognition problem by offering a robust and flexible combined representations of both the boundary and interior regions of logs. Furthermore, the proposed approach represents an example of how to integrate the emerging technologies to help the sawmill industry achieve better lumber value and volume yield.