Pattern Recognition: initial recognition

1. pattern recognition is the process of "external information channels are sensory organs and converted into meaningful sensory experiences. Sensory experience can be understood as a pattern, and the process of "converting to meaningful sensory experience" is the pattern "recognition process ". Pattern recognition problems are usually identified or classified.

2. The pattern recognition process is generally as follows:

(1) Information Input and data acquisition.

That is, to obtain the original data, such as the photos taken.

(2) data preprocessing.

To facilitate feature extraction, data must be preprocessed, such as noise reduction and target division.

(3) feature extraction, selection, and extraction.

Extraction: Calculate all measurable features. For example, a total of 10 features are extracted from a data set, including x1, x2, X3, X4, X5, X6, X7, X8, x9, and x10.

Select: select a part of the extracted features as the modeling data to avoid the feature dimension being too large. For example, select only five features for modeling x1, x2, X3, X4, X5, vector x = (x1, x2, X3, X4, X5) is called a feature vector. All combinations constitute the feature space.

Extraction: sometimes some transformation technology can be used to obtain a number of less comprehensive features for classification, known as feature dimension compression, and also become feature extraction. For example, for F and G transformations, Y1 = f (x1, x2, X3, X4, X5), y2 = g (x1, x2, X3, X4, X5 ), in this way, the X feature space is transformed to the Y feature space, and the feature dimension is changed from 5 to 2.

(4) discriminative Classification

Multiple Identification Methods

3. pattern recognition can be divided into the following categories based on theory:

(1) Statistical Pattern Recognition: it is the main theory of pattern recognition based on probability statistics.

(2) Syntactic Pattern Recognition: Based on the formal language theory, the development is slow

(3) Fuzzy Pattern Recognition: Based on the affiliation in fuzzy mathematics, it develops rapidly and conforms to the process of human understanding of things. The material world is not absolutely bounded, there is a fuzzy transition process.

(4) neural network pattern recognition: A combination of Artificial Neural Networks and pattern recognition, simulating the operating characteristics of human brain nerve cells. In recent years, deep neural networks (dnn) have developed rapidly and are a new trend. In addition, Google's super brain program is based on dnn.

4. pattern recognition can be divided into two categories according to the implementation method:

(1) Supervised Classification

Sufficient prior knowledge is required, that is, the discriminant function is determined based on the distribution of their feature vectors based on the training samples of known classes, and then the unknown pattern is identified.

(2) unsupervised classification

No prior knowledge is required, which is generally used in the absence of prior knowledge. Generally, clustering analysis is used based on the idea of "taking things together. _ {3} ^ {1} {e}