深度置信网络模型的机载多光谱数据罂粟识别

Poppy Detection in Airborne Multispectral Data Based on Deep Belief Network

针对传统识别算法在罂粟植物地块识别上精度不足的问题,提出了基于深度置信网络模型的机载多光谱数据罂粟识别算法,采用模拟人脑多层结构的方式,可以对数据自动地进行特征提取,挖掘内在联系,建立更准确的识别模型;同时将随机隐退过程引入到罂粟识别的深度网络中,避免了传统神经网络因为随机初始化而陷入局部最优解的情况。无人机航拍数据的实验表明,在小样本罂粟训练集的情况下,与支持向量机和传统神经网络方法相比,基于随机隐退的深度置信网络模型可取得更好的识别结果。

To solve the problem of low detection precision on poppy planting blocks, we propose a poppy detection algorithm based on DBN with dropout, which avoids the situation that traditional neural networks caught in the local optimal solution because of random initialization. This algorithm allows computational models that are composed of multiple processing layers to learn representations of data with multiple levels of abstraction. Aerial multispectral data from unmanned aerial vehicles was used for evaluation. The experimental results show the effectiveness of the proposed method with small training sample set in comparison with support vector machine and traditional neural network methods.