Neural network classification
After the feature extraction and selection are completed, the classifier is used to classify the image target. In this paper, the BP network in the neural network is used for classification. When designing a neural network structure, the number of layers of the network, the number of neurons in each layer, and the activation function of each layer must be considered. The standard BP neural network is divided into 3 layers, namely the input layer, the hidden layer and the output layer.
According to this layering, because the input of the neural network is the output of feature selection, it is selected from 1 to 9, and the number of neurons in the input layer of the neural network should be selected from 1 to 9. The number of neurons in the output layer can be the number of target categories to be classified, or it can be encoded according to the category to meet the output of m categories. An output unit is used, that is, the output of two categories 
= 1 output unit. When the output is 1, it can be determined as the first category, and when the output is 2, it is the second category. The selection of the number of hidden layer units is a complex problem, that is, the number of hidden layer units is directly related to the requirements of the problem and the number of input and output units.
For the BP network used for classification, because the input and output of the hidden layer unit are monotonically rising non-linear functions, the number of hidden layer units may be too small to train, or the network is not strong enough to recognize The samples we see have poor fault tolerance, but too many hidden layer units make the learning process too long, and the error is not necessarily optimal, so there is an optimal number of hidden layer units. The solution is given below
Analysis of classification results Figure 2 is the original remote sensing image. Through manual interpretation, it is determined that the features in the image are divided into roads, residential areas and high-rise buildings. Each class selects 1,000 learning samples, a total of 3 000 learning samples. Three of the nine feature quantities are selected, so the number of nodes in the input layer of the network is three. After calculation, the hidden layer takes five nodes.
Nantong Boxin Electronic Technology Co., Ltd. , https://www.bosencontrols.com