基于卷积神经网络的航拍影像信息识别方法

Aerial photographic information recognition based on convolutional neural network

利用TensorFlow机器学习平台改进基于区域卷积神经网络信息识别模型,实现基于无人机影像的目标信息的识别。以GPU为实验环境,通过卷积神经网络模型中自由参数的设置,增加训练样本的多样性,优化改进YOLO模型,调配出最优化参数。实验采用验证数据集上的植被图片对模型进行训练,准确度达到88%,将模型用来处理无人机影像数据集达到预期效果,实验证明了本方法的有效性。

A tensorflow machine learning platform was used to improve the information recognition model based on regional convolutional neural network,and to realize target information recognition based on aerial photography images.Taking the GPU as the experimental environment,the diversity of training samples was increased by setting free parameters in the convolutional neural network model.The Yolo method of information recognition was studied and the parameters were optimized.In the experiments,the vegetation images on the verification data set were used to pre-train the model,and the accuracy reached 88%.The model was used to process the aerial photography data set and achieve desired effects.This method was validated by the experiments of shooting aerial photographic images.