田间作物杂草识别的光谱通道选择

Spectral channels selection of weed identification in the field

利用地面成像光谱数据开展针对田间作物杂草识别的高光谱波段通道选择与宽波段通道选择研究。在高光谱波段通道选择方面,利用Bhattacharyya距离法筛选出对不同种类植物生化特征等敏感的通道,在优选出10个通道的情况下总体分类精度可达344维高光谱特征下总体分类精度的96%。在宽波段通道选择方面,从光谱特性的角度出发,利用模拟出的3~4个宽波段通道的光谱特征时识别精度在80%左右,利用模拟出的5~7个宽波段通道的光谱特征时识别精度可在85%以上。该研究结论可为田间作物杂草识别的传感器类型选择及光谱通道选择提供参考。

A Weed identification system in the field was investigated by hyperspectral channel selection and wide-band channel selection using ground imaging spectrometer data.Sensitive hyperspectral channels to biochemistry character were selected by using the Bhattacharyya distance method.The classification accuracy can achieve 96% by utilizing 344 dimension spectrometer data with 10 selected hyperspectral channels.The classification accuracy can achieve around 80%by using 3~4stimulated wide-band channels and can achieve more than 85%by using 5~7stimulated wide-band channels.These results provide references for sensor and channel selection in weed identification in the field.