Summary form only given. Current segmentation and classification algorithms are not sufficiently robust to provide reliable real-time sensor-based vehicle guidance for intelligent vehicle highway systems. An alternati...
详细信息
Summary form only given. Current segmentation and classification algorithms are not sufficiently robust to provide reliable real-time sensor-based vehicle guidance for intelligent vehicle highway systems. An alternative technique based on neural networks was developed for scene classification as these algorithms can handle missing and fuzzy data. The backpropagation algorithm was successfully used in conjunction with new activation functions for classification of roads, lane markers, shadows, grass, and edge transitions in real-highway scenes. 131 subregions of size 3*3 with known classes such as roads, lane markers, shadows, grass, and low- and high-edge transitions were extracted from 15 training images. Two different neural network architectures based on image and edge data were defined. The neural network was then trained for learning the characteristics of desired classes. After convergence of the training phase was completed, the test images were correctly classified into the desired classes. The training time of 2.14 h was significantly lower than that of days to a week, as reported by other researchers for similar applications.< >
The authors examine the usefulness of the feedforward neural network as a controller. For illustrative purposes, the authors consider the case of controlling two-dimensional linear systems. Observations are then made ...
详细信息
The authors examine the usefulness of the feedforward neural network as a controller. For illustrative purposes, the authors consider the case of controlling two-dimensional linear systems. Observations are then made which generalize to higher dimensions and nonlinear systems. Examples are provided to verify the results. In particular, a classical power system stabilizer is examined to demonstrate the feasibility of using a neural controller.< >
A novel approach to the problem of speaker-independent vowel recognition is presented. A novel neural architecture and learning algorithm called neural tree networks (NTNs) are developed. This network uses a tree stru...
详细信息
A novel approach to the problem of speaker-independent vowel recognition is presented. A novel neural architecture and learning algorithm called neural tree networks (NTNs) are developed. This network uses a tree structure with a neural network at each tree node. The NTN architecture offers a very efficient hardware implementation as compared to MLPs (multilayer perceptrons). The NTN algorithm grows the neurons while learning as opposed to backpropagation, for which the number of neurons must be known before learning can begin. The proposed algorithm is guaranteed to converge on the training set whereas backpropagation can get stuck in local minima. Simulation results on a speaker-independent vowel-recognition task are presented which show that the new method is superior to both the MLP and decision tree methods.< >
We present two algorithms for unlearning an already stored pattern in feed- forward neural networks. The proposed algorithms are modifications of the backpropagation method. The method will find applications whenever ...
详细信息
We present two algorithms for unlearning an already stored pattern in feed- forward neural networks. The proposed algorithms are modifications of the backpropagation method. The method will find applications whenever the training set has to be updated frequently. The proposed algorithms have the propery that the retrieval of other pattens are not affected due to the selective unlearning.
The automatic classification of acoustic emission signals is discussed. A multilayer heterogeneous network was designed to improve the performance of the structure and reduce training time. The network consists of two...
详细信息
The automatic classification of acoustic emission signals is discussed. A multilayer heterogeneous network was designed to improve the performance of the structure and reduce training time. The network consists of two basic subnetworks, a compression subnetwork with a generalized Hebbian algorithm, and a classification subnetwork with a backpropagation algorithm. Feature extraction and data compression are accomplished by the compression network first, and then classification is done by the classification network using the compressed data. The signal preprocessing provided by the Hebbian algorithm contributes to the optimal linear data reconstruction with maximal variance and minimal error. Classification performances using both compressed and raw data are compared. Significant reductions in the size of the network and in the training time were achieved in a simulation. The network structure, generalization capability, and classification accuracy are all discussed.< >
A study is made on the application of the artificial neural network (ANN) method to forecast the short-term load for a large power system. The load has two distinct patterns: weekday and weekend-day patterns. The week...
详细信息
A study is made on the application of the artificial neural network (ANN) method to forecast the short-term load for a large power system. The load has two distinct patterns: weekday and weekend-day patterns. The weekend-day pattern include Saturday, Sunday, and Monday loads. Three different ANN models are proposed, including two feedforward neural networks and one recurrent neural network. Inputs to the ANN are past loads and the output is the predicted load for a given day. The standard deviation and percent error of each model are compared.< >
The authors describe experiments in which the rescaling back-propagation learning algorithm was used to learn sets of linear filters for the task of determining the orientation and location of edges to subpixel accura...
详细信息
The authors describe experiments in which the rescaling back-propagation learning algorithm was used to learn sets of linear filters for the task of determining the orientation and location of edges to subpixel accuracy. A model of edge formation was used to generate input-output pairs for each iteration of the training process. The desired output included determining the interpolated location and orientation of the edge.< >
A series of experiments with the cascade-correlation algorithm (CCA) and some of its variants on a number of real-world pattern classification tasks are described. Some of the experiments investigated the effect of di...
详细信息
A series of experiments with the cascade-correlation algorithm (CCA) and some of its variants on a number of real-world pattern classification tasks are described. Some of the experiments investigated the effect of different design parameters on the performance of the CCA. Parameter settings that consistently yield good performance on different data sets were identified. The performance of the CCA is compared with that of the backpropagation algorithm and the perceptron algorithm. Preliminary results obtained from some variants of CCA and some directions for future work with CCA-like neural network learning methods are discussed.< >
暂无评论