ANNS in Brain Signal Process

1. Feature extraction, classification, and pattern recognition: ANNs here serve mainly as non-linear classifiers. The inputs are preprocessed so as to form a feature space. ANNs are used to categorize the collected data into distinct classes. In other cases, inputs are not subjected to preprocessing but are given directly to an ANN to extract features of interest from the data.

2. Adaptive filtering and control: ANNs here operate within closed loop systems to process changing inputs, adapting their weights “on the fly” to filter out unwanted parts of the input (adaptive filtering), or mapping their outputs to parameters used in online control (adaptive control).

3. Linear or nonlinear mapping: Here ANNs are used to transform inputs to outputs of a desired form. For example, an ANN might remap its rectangular input data coordinates to circular or more general coordinate systems.

4. Modeling: ANNs can be thought of as function generators that generate an output data series based on a learned function or data model. ANNs with two layers of trainable weights have been proven capable of approximating any nonlinear function.

5. Signal separation and deconvolution: These ANNs separate their input signals into the weighted sum or convolution of a number of underlying sources using assumptions about the nature of the sources or of their interrelationships (e.g., their independence).

6. Texture analysis and image segmentation: Image texture analysis is becoming increasingly important in image segmentation, recognition and understanding. ANNs are being used to learn spatial or spatial-frequency texture features and, accordingly, to categorize images or to separate an image into sub images (image segmentation).

7. Edge detection: In an image, an edge or boundary between two objects can be mapped to a dark band between two lighter areas (objects). By using the properties of intensity discontinuity, ANNs can be trained to “recognize” these dark bands as edges, or can learn to "draw" such edges based on contrast and other information.