无人机深度成像背景下农田伴生杂草精准分类  

Accurate Classification of Associated Weeds in Farmland under Background of UAV Deep Imaging

作　　者：秦浩 石元伍 QIN Hao;SHI Yuan-wu(College of Industrial Design,Hubei University of Technology,Wuhan Hubei 430068,China;Wuhan Textile University,Wuhan Hubei 430074,China)

机构地区：[1]湖北工业大学工业设计学院,湖北武汉430068 [2]武汉纺织大学,湖北武汉430074

出　　处：《计算机仿真》2024年第9期510-514,共5页Computer Simulation

摘　　要：农田图像中存在各种背景干扰,且在不同生长周期和环境条件下杂草的外观差异也很大,导致在分类过程中,难以准确识别杂草,降低其分类精度,因此,提出无人机深度成像下农田伴生杂草精准分类算法。该方法基于无人机获取农田伴生杂草深度图像,并对图像展开去噪以及清晰度增强;根据处理结果对农田深度图像实施色调、饱和度、明度(Hue,Saturation,Value,HSV)分解,利用改进的二维局部熵算法,结合人眼视觉,完成图像对比度、饱和度以及轮廓信息的密度计算,建立图像显著性矩阵;提取图像中感兴趣区域实施背景分割,结合改进尺度不变特征完成伴生农田特征提取;利用图像分类器对提取特征实施特征分类处理,从而完成农田伴生杂草精准分类。实验结果表明,利用该方法开展农田伴生杂草分类时,去噪效果好,目标轮廓提取完整,且分类准确率在99%以上。There are various background interferences in farmland images,and the appearance of weeds varies greatly under different growth cycles and environmental conditions,which makes it difficult to accurately identify weeds during the classification process and reduces their classification accuracy.Therefore,a precise classification algorithm for associated weeds in farmland under unmanned aerial vehicle deep imaging is proposed.This method is based on unmanned aerial vehicles to obtain depth images of associated weeds in farmland,and then to denoised and enhanced the clarity of the images.Based on the processing results,Hue,Saturation,Value(HSV) decomposition is performed on the depth image of farmland.Moreover,an improved two-dimensional local entropy algorithm is combined with human vision to calculate the density of image contrast,saturation,and contour information,thus constructing an image saliency matrix.Furthermore,the region of interest in the image is extracted,and then the background is segmented.Based on the scale invariant features,the associated farmland features can be extracted.Finally,the image classifier is used to classify the extracted features,thereby achieving the precise classification of associated weeds in farmland.The experimental results show that this method has good denoising effect and complete contour extraction.The classification accuracy is more than 99%.

关 键 词：无人机深度成像 农田伴生杂草图像 图像去噪 分类方法 特征提取 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]