Development of an integrated sugarcane quality management system using combined use of optical sensors

2022年，我们在日本冲绳的南大东岛进行了甘蔗生长或成熟期（2022年12月-2022年12月）的试验。利用便携式可见-近红外光谱仪对田间甘蔗茎秆的糖度和吸收光谱进行了测定。此外，我们收集了甘蔗叶的多光谱图像，使用配备了多光谱相机的无人驾驶飞行器。我们通过检查便携式可见-近红外光谱仪测量的糖质量与植被指数之间的相关性，评估了多光谱图像在估计糖质量方面的有效性。此外，我们还调查了甘蔗皮对南大东的影响（2023年1月至3月）。我们扫描了正常的甘蔗茎和来自不同田地和品种的去皮甘蔗茎，然后收获甘蔗并将其带回实验室测量糖质量并开发新的校准模型。在使用多元线性回归方法建立模型后，我们发现去皮甘蔗光谱比未去皮光谱显示出更低的噪声，未去皮光谱具有由甘蔗蜡引起的更高的噪声。结果表明，去皮甘蔗光谱模型比未去皮甘蔗光谱模型提供了更准确的结果，因为去皮甘蔗光谱模型中相关系数最高的前两个波长包含糖含量吸收带（900 - 910 nm），而正常甘蔗茎光谱模型必须选择至少四个波长才能找到糖含量吸收带。

In 2022, we conducted experiments on Minamidaito Island in Okinawa, Japan during the growth or ripening stage of sugarcane (October-December 2022). We measured sugar quality and absorbance spectra of sugarcane stalks in the sugarcane field using a portable Vis-NIR spectrometer. Additionally, we collected multispectral images of sugarcane leaves using an unmanned aerial vehicle equipped with a multispectral camera. We evaluated the effectiveness of the multispectral images in estimating sugar quality by checking the correlation between sugar quality measured by the portable Vis-NIR spectrometer and vegetation indices.Furthermore, we investigated the effect of sugarcane skin on Minamidaito (Jan-Mar 2023). We scanned normal sugarcane stalks and peeled sugarcane stalks from different fields and varieties, then harvested the sugarcane and brought it back to the laboratory to measure sugar quality and develop a new calibration model. After developing models using the multiple linear regression method, we found that the peeled sugarcane spectra showed lower noise than the non-peeled spectra, which had higher noise caused by sugarcane wax. The results reveal that the peeled spectra model provides more accurate results than the non-peeled sugarcane spectra model, as the top two wavelengths with the highest correlation coefficient of the peeled sugarcane spectra model contain the sugar content absorption band (900-910 nanometers), whereas the normal sugarcane stalk spectra model had to select at least four wavelengths to find the sugar content absorption band.