Three research prototype speech recognition systems are described, all of which use recently developed methods from artificial intelligence (specifically support vector machines, dynamic Bayesian networks, and maximum entropy classification) in order to implement, in the form of an automatic speech recognizer, current theories of human speech perception and phonology (specifically landmark-based speech perception, nonlinear phonology, and articulatory phonology). All three systems begin with a high-dimensional multiframe acoustic-to-distinctive feature transformation, implemented using support vector machines trained to detect and classify acoustic phonetic landmarks. Distinctive feature probabilities estimated by the support vector machines are then integrated using one of three pronunciation models: a dynamic programming algorithm that assumes canonical pronunciation of each word, a dynamic Bayesian network implementation of articulatory phonology, or a discriminative pronunciation model trained using the methods of maximum entropy classification. Log probability scores computed by these models are then combined, using log-linear combination, with other word scores available in the lattice output of a first-pass recognizer, and the resulting combination score is used to compute a second-pass speech recognition output.

The quality of the wood is determined by the number of defects and its distribution. In a piece of timber, the most common type of imperfection is called knot that decreases the strength of the wood. Manual selection and classification process of knots is tedious and time consuming job. An automatic sensing machine is able to inspect wood automatically and correctly identify the defects it possess, and its effect on the quality of the final product. In this paper, it is proposed to detect and classify the knots in timber boards. The image of knots is pre processed using Hilbert transform and Gabor filters. The features obtained from pre processing, is classified using data mining techniques and compared with bagging technique.

Plant diseases are a severe cause of crop losses in the agriculture globally. Detection of diseases in plants is difficult and challenging due to the lack of expert knowledge. Deep learning-based models provide promising ways to identify plant diseases using leaf images. However, need of larger training sets, computational complexity, and overfitting, etc. are the major issues with these techniques that still need to be addressed. In this work, a convolutional neural network (CNN) is developed that consists of smaller number of layers leading to lower computational burden. Some augmentation techniques such as shift, shear, scaling, zoom, and flipping are applied to generate additional samples increasing the training set without actually capturing more images. The CNN model is trained for apple crop using a publicly available dataset PlantVillage to identify Scab, Black rot, and Cedar rust diseases in apple leaves. The rigorous experimental results revealed that the proposed model is well fit to identify apple leaf diseases and achieves 98% classification accuracy. It is also evident from the results that it needs lesser amount of storage and takes smaller execution time than several existing deep CNN models. Although, there exist several CNN models for crop disease detection with comparable accuracy, but the proposed model needs lower storage and computational resources. Therefore, it is highly suitable for deploying in handheld devices.

No abstract

No abstract