为了快速感知并分析田间作物生长状况, 采用先进的半导体镀膜工艺的光谱成像传感器, 研究镀膜型光谱成像数据的提取与叶绿素含量分布式检测的方法。 实验采用基于镀膜原理的IMEC 5×5成像单元式多光谱相机, 对47株苗期玉米植株的冠层进行拍摄, 获取673～951 nm范围内的25个波长的光谱图像。 利用SPAD-520叶绿素仪非破坏性地测量叶绿素含量指标, 每株玉米冠层叶片设置2～3个采样点, 每点测量3次取平均, 共计251个样本数据; 同时使用ASD Handheld2型光谱仪采集相应位置区域的反射率曲线, 以对比分析镀膜型光谱成像传感器提取玉米植株冠层叶片反射率曲线的特性。 首先, 在分析镀膜型光谱成像传感器的成像原理的基础上, 通过对原始图像的拆分和重组分别提取成像单元中相同波段的像素灰度值, 并利用相同波段的像素灰度值重构单波段光谱图像, 获取各波段光谱图像。 其次, 利用4灰度级标准板建立图像灰度值和灰度板反射率之间的线性反演公式, 对提取的反射率进行校准。 然后, 为了准确分割出玉米植株冠层, 提出了大津算法(OTSU)和霍夫圆变换组合的玉米植株冠层图像二次分割方法, 分别剔除图像中土壤和培养盆背景的干扰。 最后, 利用马氏距离算法剔除异常样本数据, 利用SPXY (sample set partitioning based on joint X-Y distance)算法划分建模集和验证集, 采用偏最小二乘回归法(PLSR)建立玉米植株叶绿素含量指标诊断模型, 并绘制其分布伪彩色图用于分析叶绿素含量空间分布特征。 研究结果表明, ①对25波段多光谱图像提取和反射率线性校准拟合模型决定系数均达到0.99以上。 分析校准前和校准后与ASD光谱仪测量反射率曲线, 镀膜型成像传感器获取玉米冠层反射光谱总体与ASD采集反射率体现的光谱特征一致, 且校正后数据比校正前与ASD光谱反射率的一致性得到了提升。 ②建立初次OTSU分割算法和基于霍夫圆变换识别的二次分割算法, 可以有效剔除玉米植株光谱图像中的土壤和培养盆背景噪声的干扰。 ③叶绿素含量指标PLSR诊断模型建模集R2c为0.545 1, 验证集R2v为0.472 6。 玉米作物冠层叶绿素分布可视化图可以直观反映叶绿素含量分布与生长动态情况。 通过对镀膜型光谱成像传感器应用方法的研究, 为后续玉米植株叶绿素动态快速检测奠定基础和提供技术支持。

In order to quickly analyze the growth of the crop in the field, the spectral imaging sensor was used to detect the chlorophyll content of the maize canopy. The images of 47 maize plants were photographed using an IMEC 5×5 imaging unit multispectral camera. The camera was designed based on the coating principle to obtain spectral images of 25 wavelengths in the range of 673～951 nm. At the same time, the chlorophyll content was measured by SPAD-520 device. There were 2～3 sampling points in each leaf, and they were measured 3 times at each point, so that 251 sample data were collected. The multi-spectral images were processed. Firstly, the gray pixel values of the same band in the imaging unit were extracted based on the principle of the coating spectral imaging sensor. The extraction methods included image splitting and recombination, in which 25 images were extracted from the original image. Secondly, a linear inversion formula between the gray value of multi-spectral images and the gray plate standard reflectance was established. The gray plate standard was made up of 4 gray level standard plates. Thirdly, the image segmentation algorithm was established to reduce the background influence in maize canopy images, in which the OTSU algorithm was used to eliminate the interference of the soil and the Hough circle transform algorithm was used to eliminate the interference of the flowerpot. Lastly, the study used the Mahala Nobis distance algorithm to eliminate abnormal sample data. According to the proportion of 2∶1, the total samples were divided into calibration set (170 samples) and validation set (73 samples) by SPXY (Sample set partitioning based on joint X-Y distance) algorithm. The partial least squares regression (PLSR) model was established to detect the chlorophyll content of the maize canopy. In general, the results of spectral bands reflectance linear calibration fitting model were above 0.99. The corrected data was consistent with the ASD spectral reflectance before calibration. The interference of soil and flowerpot background noise in the multi-spectral images were removed by the image segmentation algorithm. The calibration accuracy of PLSR model was 0.545 1, and the validation accuracy was 0.472 6. And then the chlorophyll content of each pixel in the maize canopy images could be calculated by the PLSR modeling result. As a result, the chlorophyll content distribution could be visually analyzed and indicated the growth status. The study could provide technical and application support for the chlorophyll distribution of field maize plants.