| Summary: | 博士 === 國立成功大學 === 電機工程學系 === 87 === In recent years, work on neural networks has a rapid growth in both theory and practical applications because of its learning ability, parallel processing and hardware implementability. Among many network architectures, spatiotemporal neural network specializes in vision, speech recognition, signal processing and motor control through the incorporation of time into its operation.
In this dissertation, we first examine developed architectures and models of spatiotemporal neural network and then propose a new spatiotemporal network capable of recognizing one-dimensional periodic signals. Based on the proposed network, we construct a recognition system for single object and partially occluded objects. The system consists of two major components: a feature extraction process and a spatiotemporal modular neural network. The former is made up of a sequence of preprocessing techniques including thresholding, boundary extraction, Gaussian filtering and split-and-merge algorithm to generate features that will represent the objects to be recognized. These acquired features are invariant to rotation, translation and scaling and can serve as input to the spatiotemporal network which utilizes the concept of tap-delay to account for spatial correlation between consecutive input features. A shape perceiver is designed into the network to extract continued parts of an object, and also to enable the inclusion of each object''s unique characteristics into the system. Traditional neural network approaches for recognizing partially occluded objects have encountered significant problems because of the incomplete boundaries of the objects. In our approach, by creatively installing tap-delays, the system can escape this limitation. Experimental results show that the proposed system can produce satisfactory results in efficiently and effectively recognizing partially occluded objects. Furthermore, intrinsic to this system is the ease by which it can be realized through parallel implementation, thus minimizing the tremendous time spent in matching object contours stored in a database model, as is the case in conventional recognition systems.
We also propose a modified probabilistic neural network (MPNN) combined with a modified avalanche neural network to classify liver tumors into hepatoma and hemageoma. Temporal information, derived from fractal geometry, and spatial feature descriptors, generated by the spatial gray-level co-occurrence matrix, are fed into the MPNN to come to a conclusion. The proposed system was evaluated by 30 liver cases and shown to be efficient and very effective.
|
|---|
