Fine-grained region classification method and system based on convolutional neural network

Abstract

The invention relates to a fine-grained region classification method and system based on a convolutional neural network, and the method comprises the steps: S1, inputting a UJIIndoorLoc data set into an SAE-1D Resnet10 convolutional neural network, and obtaining a building classification and layering result; s2, calculating an error between a classification result and a true value by using a quadratic cost function; s3, segmenting the data set based on the CSI into bins with the size of w; s4, performing CNN state reasoning model training, and outputting a state label; s5, inputting the CSI amplitude of the state label into a CNN state inference model for training, and outputting probability distribution of fine-grained region classification; s6, calculating a concentration ratio by using the maximum probability and the probability entropy, and obtaining a sample confidence coefficient; and S7, dynamically adjusting an active segmentation algorithm. According to the method, the optimal state segmentation threshold determined by artificial subjective observation and experience is replaced by a mode of dynamically adjusting the state segmentation algorithm by utilizing CSI active window segmentation and CNN model training classification state labels, and a feedback mechanism, so that the sample confidence is higher.

Description

technical field [0001] The invention relates to the technical field of multi-building and multi-floor area classification, in particular to a convolutional neural network-based fine-grained area classification method and system. Background technique [0002] In recent years, with the continuous advancement of communication technology, the demand for high-precision classification based on fine-grained areas of buildings and floors has been increasing in emergency rescue, item search, path planning, etc. However, the Global Positioning System (Global Positioning System, GPS) and Beidou satellite The BeiDou Navigation Satellite System (BDS) is increasingly unable to meet the needs of indoor fine-grained area classification. Therefore, how to improve the accuracy of indoor fine-grained area classification has become an urgent research topic. [0003] The technical means of area classification usually include computer vision, inertial sensors, UWB and wearable devices, etc. Alth...

AI Technical Summary

Problems solved by technology

Traditional machine learning algorithms are easily affected by the environment, and the classification error is large under non-line-of-sight conditions, and they are not suitable for area classification with multiple floors and buildings

By training the offline CSI fingerprint library and RSS fingerprint library, the neural network can perform coarse-grained area classification of multi-floor, multi-building, and multi-room, but it cannot be subdivided into small areas.

Images