基于无人机可见光影像的高原丘陵盆地区山药植株识别

采用四旋翼无人机采集特色作物山药种植区影像，针对山药种植时间和管护水平差异导致的植株生长差异化特征，通过筛选红绿比值指数（RGRI）、过绿指数（ExG）和过绿减过红指数（ExG-ExR）等颜色指数获取山药植株最优颜色指数计算方法；以高斯高通滤波（GHPF）增强和保留山药植株高频信息，抑制杂草、玉米植株的噪声；结合田间测量山药植株数据，通过人机交互解译方法对图像滤波增强后的影像进行灰度分割和提取山药植株信息。结果表明：1）样区总体提取精度为91.14%，样区A、B、C的提取精度分别为90.94%、91.96%、90.81%，精度验证完整性为93.16%；2）随着山药植株多株连体生长复杂程度的增强，过绿指数具有的土壤与植被的分离性仍优于红绿比值指数和过绿减过红指数；3）使用高斯高通滤波能够有效消除杂草、玉米植株的影响，减小山药植株多株连接生长所产生的干扰；4）高斯高通滤波卷积核大小79×79适用于不同时相和不同生长情况的山药植株影像处理，针对不同的山药植株生长情况和不同时相的可见光影像，需调整灰度分割的阈值参数t，单株山药植株面积S由田间测量确定。该方法以快速灵活、低成本的方式识别和计算不同生长状态的山药植株数目，适用于喀斯特山区的精准农业监测研究和现代农业生产活动。

Identification of Yam Plants in Karst Plateau Hill Basin Based on Visible Light Images of an Unmanned Aerial Vehicle

The images of yam plantation areas were collected by the four-rotor unmanned aerial vehicle (UAV). Aiming at different characteristics of the yam plant growth caused by diverse planting times and management levels, the optimal color index calculation method was obtained by screening Red-Green Ratio Index (RGRI), Excess Green Index (ExG), Excess Green Minus Excess Red Index (ExG-ExR), and other color indexes. Gaussian High-Pass Filter (GHPF) was used to enhance and retain high-frequency information of yam plants and inhibit the noise from weeds and corn plants. Combined with field measurement of yam plants, grayscale segmentation and extraction of yam plant information were performed on the data after image filtering and image enhancement via a human-computer interaction interpretation method. The results show that: 1) the method has high yam plant identification and plant number calculation accuracy. The overall average extraction accuracy of the sample areas is 91.14%, and the extraction accuracy of sample area A, sample area B, and sample area C is 90.94%, 91.96%, and 90.81%, respectively. The extraction accuracy of the image for accuracy verification integrity is 93.16%. 2) With the increasing complexity of multiple conjoined growth of yam plants, the soil and vegetation separation of ExG is still superior to that of RGRI and ExG-ExR. Soil and vegetation separation gradually decreases while calculating RGRI, while the plant pattern spots gradually expand into one pattern spot. In the calculation of ExG-ExR, the soil and vegetation separation decreases with an increase in the complexity of multiple conjoined plants. In addition, the calculation method of ExG-ExR considerably contributes to the DN (Digital Number) value of weeds and corn plants. The DN value of weeds and corn is similar to that of yam plants, which is not conducive to the identification of the latter. 3) GHPF method has a good enhancement effect on the high-frequency information of the yam plants; the mean, range, and standard deviation, respectively, transform from 3.2044 to -0.0198, 181 to 121, and 25.5886 to 15.9735, and the weeds and corn plants are effectively inhibited. Thus, the use of GHPF can effectively eliminate the effects of weeds and corn plants, and reduce the interference caused by the multiple conjoined growth of yam plants. The yam plant identification and extraction method using ExG, GHPF, and Individual Crop Area Method can effectively eliminate the influence of weeds and reduce the interference caused by the growth of multiple yam plants; 4) The nucleus size 79×79 of GHPF is suitable for image processing of yam plants at different time phases and under different growth conditions. For visible light images of yam plants under different growth conditions and different growth phases, it is necessary to adjust the threshold parameter t of gray scale segmentation, and determine the area S of individual yam plant by field measurements. The method identifies and calculates the number of yam plants in various stages of growth states in a fast, flexible and low-cost manner, which is suitable for precision agricultural monitoring research and modern agricultural production activities in karst mountainous areas.