Shigetoshi Shiotani

Application of image analysis with neural network for plant somatic embryo culture

Abstract A method of classifying celery embryos and nonembryos using image analysis with a neural network to decide the time for transfer to the next culture stage in plant somatic embryo culture is presented. Since the image database of celery cells is vast, four key input parameters (area, ratio of length to width, circularity and distance dispersion) were selected. Among these four parameters, use of the first three was found to be satisfactory for classification between embryos and nonembryos. Using the three parameters, the trained neural network was also able to classify globular, heart- and torpedo-shaped embryos at a level comparable with estimation by a human expert. By using the trained neural network, the number of plantlets that would be formed after the second regeneration culture of 14 d could be successfully predicted from number of heart- and torpedo-shaped embryos at the end of the first regeneration stage.

A neural network architecture for incremental learning

Abstract Artificial neural networks have been used as a tool for category classification. The neural network can correctly classify patterns which have already been trained. However, sometimes the neural network erroneously classifies patterns which have never been trained. The neural network must learn again to correct the errors. In this learning, the multi-layered perceptron (MLP) must learn new patterns and old patterns. The new pattern is the pattern which the MLP cannot classify correctly and the old pattern is the pattern which the MLP has already learned. So, the MLP is ineffective in computing cost due to learning the old patterns. The adaptive resonance theory (ART) model can memorize the new patterns without learning the old patterns due to incremental learning. However, it has problems with classification ability. This paper proposes a neural network architecture for incremental learning. This neural network is called ‘Neural network based on Distance between Patterns’ (NDP). The NDP has a two-layered hierarchical structure and many neurons of the radial basis function in the output layer. The NDP performs incremental learning which increases neurons in the output layer and varies the center and the gradient of the radial basis function. So, the NDP can memorize the new patterns without learning the old patterns and has superior classification ability. The NDP differs from conventional radial basis function neural networks in the area of incremental learning. In addition, this paper shows the effectiveness of the NDP in experiments on image recognition.

A new neuron model for additional learning

A novel neuron model called the new neural network (NNN) is proposed. It is shown that the NNN can learn and memorize additionally and recognize unlearned patterns by its generalization for two simulations on recognition. The NNN can recognize unlearned patterns more efficiently than backpropagation by the evaluation function in which the similarity is considered. The NNN has two excellent abilities: additional learning and superior generalization.<<ETX>>

RECOGNITION AND COUNTING METHOD OF MAMMALIAN CELLS ON MICRO·CARRIER USING IMAGE PROCESSING AND NEURAL NETWORK

This paper presents a method for recognizing mammalian cells (fibroblast) on micro-carrier (bead; bead is sphere-shaped and transparent) using a technique of image processing and neural network. The cell culture on micro-carriers is one of the efficient cell culture methods. It is an important task to monitor optimal culture conditions such as the growth rate of cells on a bead. A system for counting the number of cells will provide an automatic production of cells. Previous studies presented the counting algorithms using image segmentation, knowledge data base systems and fuzzy inference. These algorithms are incomplete, because they cannot recognize adjoining cells and overlapping cells which are on the top and the bottom part of a bead. In this paper, we propose a new cell counting system, which is developed using image processing and neural network and considers the nucleoli in the cell. Experimental results show the realization of counting overlapping cells and adjoining cells.

Recognition system by neural network for additional learning

The authors propose a neural network (NN) model for pattern recognition which can learn new patterns without losing patterns memorized in the past. This model is called a neural network based on distance between patterns (NDP). The NDP uses the radial basis function (RBF) as the response function in place of the sigmoid function. The response function is a smooth function similar to the Gaussian base function and the probability density function. The response function learns patterns faster than the multilayered perceptron (MLP) as well as other RBF NNs. The most salient architectural feature of the NDP is self-organization by adding nodes in the output layer one by one. The NDP also varies the curve of the response function by tuning the center and the width of the response function, and separates the input space into regions for each category appropriately.