基于热红外图像处理技术的农作物冠层识别方法研究

为解决农作物冠层热红外图像边缘灰度级分布不均且噪声较大，而传统图像分割方法难以实现其目标区域有效识别的难题，以苗期红小豆冠层热红外图像为研究对象，将模糊神经网络和仿射变换有机结合，提出了基于热红外图像处理技术的农作物冠层识别模型。首先利用五层线性归一化模糊神经网络的自适应特性，选取高斯隶属度函数，自动计算冠层可见光图像识别的推理规则，有效地分割了可见光图像中的冠层区域。通过分析3种分割指标和熵，定量评价可见光图像冠层分割质量。网络迭代38次时，误差精度为0.000 952，该算法平均有效识别率为96.13%，获取可见光冠层图像的像元信息熵值范围为2.454 4~5.198 7，与标准算法所得冠层图像的像元信息熵仅相差0.245 9。然后以取得可见光图像的冠层有效区域为参考图像，采用仿射变换算法，调整优选平移、旋转、缩放等图像变换因子，配准原始热红外图像，提出了基于仿射变换的冠层热红外图像识别方法。对于初始温度范围值在16.35~19.92 ℃的农作物热红外图像，计算选取旋转幅度为1.0和缩放因子为0.9时，作为异源图像的最优配准参数，获取目标图像的最大温差为3.17 ℃，相对于原图像的平均温度值由18.711 ℃下降至17.790 ℃，进而实现了基于热红外图像处理技术的农作物冠层识别。最后以熵的互信息作为监督指标，对农作物冠层热红外图像识别方法进行评价。提出的冠层热红外图像识别方法，所获取的目标图像与初始热红外图像的平均互信息为4.368 7，标准目标图像和初始热红外图像的平均互信息为3.981 8，二者仅相差0.486 9。同时，两种冠层热红外图像的平均温度差值为0.25 ℃，高效消除了原始热红外图像的背景噪声。结果表明本研究方法的有效性和实用性，能够为应用热红外图像反映农作物生理生态信息特征指标参数提供技术借鉴。

Recognition Method for Crop Canopies Based on Thermal Infrared Image Processing Technology

In order to solve the problem that the gray level distribution of crop canopy thermal infrared image isuneven and has large noise,the traditional image segmentation method is difficult to realize the effective recognition of its target region.In this study,by the thermal infrared images of adzuki bean canopy’s in the seedling stage was taken as the research object,Combining fuzzy neural network and affine transformation,a crop canopy recognition model based on thermal infrared image processing technology was proposed.First,the adaptive characteristics of the five-layer linear normalized fuzzy neural network were used to select the Gaussian membership function to automatically calculate the inference rules for canopy visible light image recognition,effectively segmenting the canopy area in the visible light image.By analyzing three segmentation indexes and entropy,the canopy segmentation quality of visible light images was quantitatively evaluated.When the network iterates 38 times,the error precision was 0.000952,and the visible light image of the crop canopy was obtained.The average effective partition rate of the algorithm was 96.13%,and the entropy value of the image source average information was 2.4544~5.198,which was only 0.2459 different from the entropy of the canopy image obtained by the standard algorithm.Then,using the effective area of the canopy to obtain the visible light image as a reference image,the affine transformation algorithm was used to adjust the image transformation factors such as optimal translation,rotation,and scaling.To register the raw thermal infrared image.And a canopy thermal infrared image recognition method based on affine transformation was proposed.For a crop thermal infrared image with an initial temperature range of 16.35~19.92,when the rotation amplitude was 1.0 and the zoom factor was 0.9,the maximum temperature difference of the target image obtained as the optimal registration parameter of the heterogeneous image was 3.17℃.Relative The average temperature of the original image decreased from 18.711℃ to 17.790℃,and the crop canopy recognition based on thermal infrared image processing technology was realized.Finally,mutual information was used as a monitoring index to evaluate the thermal infrared image recognition method of crop canopy.In the canopy thermal infrared image recognition method proposed in this study,the average mutual information between the acquired target image and the initial thermal infrared image was 4.3687,while the average mutual information between the standard target image and the initial thermal infrared image was 3.9818,and the difference between the two was only 0.4869.At the same time,the average temperature difference between the two canopy thermal infrared images was 0.25℃,which effectively eliminates the background noise of the original thermal infrared images.The research results show that the effectiveness and practicability of this research method could provide a technical reference for the application of thermal infrared images to reflect the characteristic parameters of crop physiological and ecological information.