摘要： 利用卫星热红外异常判别技术进行地震预测的方法都是纯手工或半手工的，在分析处理海量遥感数据时具有局限性，并且传统方法对地震三要素的预测准确率不高，尤其是震中位置的预测。针对上述问题，提出一种综合震象云颜色、纹理以及浮现频率等特征来自动预测地震的方法。利用灰度共生矩阵对热红外数据进行纹理特征提取，使用BP神经网络模型训练目标神经网络，将纹理特征输入目标神经网络进行识别，提取疑似目标，同时滤掉非目标并跟踪，将疑似目标浮现频率超过5次的区域精确定位为目标出现的位置，从而实现智能化和自动化的地震预测。反演实验验证结果表明，该方法是一种震中位置预测较为准确的中短期地震预测方法。
关键词:
图像识别,
目标跟踪,
地震预测,
震象云,
灰度共生,
神经网络
Abstract: The earthquake prediction research based on interpretation technique of satellite thermal anomaly has a history of over 20 years. Previous studies are pure manual or semi-manual with many shortages in processing huge quantity remote data. Meanwhile, the traditional methods cannot give an accurate estimation on three elements of earthquakes, especially on epicenter location. In order to solve the above-mentioned problems, this paper puts forward a method based on image recognition with considering the color, texture and occurrence frequency of quake-trace cloud. An earthquake can be predicted intelligently and automatically by using automatic target detection in artificial intelligence. The entire procedure is as follows. It gets the texture features from thermal infrared data by using gray level co-occurrence, trains a target neural network by making use of BP neural network model, inputs texture features into target neural network and gets the suspected targets, filters suspected target which is undersized or oversize, tracks the remaining suspected targets, acquires the certain target by its occurrence frequency which is larger than 5, and predicts an earthquake. Experimental result shows that it is a short term earthquake prediction method with more accurate epicenter location prediction.
Key words:
image recognition,
target tracking,
earthquake prediction,
quake-trace cloud,
gray level co-occurrence,
neural network
中图分类号: