基于BP神经网络与纹理分析优化的雾检测

Optimized fog detection and classification technique based on BP artificial neural network and texture analysis

在利用NOAA AVHRR/3资料并根据雾的均匀纹理特性进行白天雾检测研究中,为了克服对整幅图像进行纹理分析存在的处理复杂和运算量大等缺点,提出了采用纹理分析方法优化细分神经网络雾检测结果的思想。通过计算神经网络检测结果中的低云和雾区连通域的灰度标准差并设定灰度标准差阈值,对神经网络检测结果中的低云区和雾区作了进一步的纹理分析优化细分。结果表明,该方法有效地提高了雾检测的准确性和可靠性。

Detection of daytime fog was studied by NOAA AVHRR/3 data according to the texture characteristics. An idea of subdividing and optimizing BP network＇s detection results by texture analysis was put forward in order to overcome the shortcomings of high complexity and large computation. Low-level clouds and fog detection results were optimized by calculating and setting the thresholds of the standard deviations of image gray in the connected domains in BP network＇ s detection results image. Results indicate that this technique effectively improves the accuracy and reliability of fog detection.