当前位置: X-MOL 学术Comput. Electron. Agric. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
Disease and pest infection detection in coconut tree through deep learning techniques
Computers and Electronics in Agriculture ( IF 8.3 ) Pub Date : 2021-01-27 , DOI: 10.1016/j.compag.2021.105986
Piyush Singh , Abhishek Verma , John Sahaya Rani Alex

The coconut palm plantation industry relies heavily on expert advice to identify and treat infections. Computer vision in deep learning technology opened up an avenue in the agriculture domain to find a solution. This study focuses on the development of an end-to-end framework to detect stem bleeding disease, leaf blight disease, and pest infection by Red palm weevil in coconut trees by applying image processing and deep learning technology. A set of hand-collected images of healthy and unhealthy coconut tree images were segmented by employing popular segmentation algorithms to easily locate the abnormal boundaries. The custom-designed deep 2D-Convolutional Neural Network (CNN) is trained to predict diseases and pest infections. Also, the state of the art Keras pre-trained CNN models VGG16, VGG19, InceptionV3, DenseNet201, MobileNet, Xception, InceptionResNetV2, and NASNetMobile were fine-tuned to classify the images either as infected or as healthy through the inductive transfer learning method. The empirical study ascertains that k-means clustering segmentation was more effective than the Thresholding and Watershed segmentation methods. Furthermore, InceptionResNetV2 and MobileNet obtained a classification accuracy of 81.48% and 82.10%, respectively, and Cohen’s Kappa values of 0.77 and 0.74, respectively. The hand-designed CNN model achieved 96.94% validation accuracy with a Kappa value of 0.91. The MobileNet model and customized 2D-CNN model were deployed in the web application through the micro-web framework Flask to automatically detect the coconut tree disease or pest infection.



中文翻译:

通过深度学习技术检测椰子树中的病虫害

可可椰子种植业严重依赖专家建议来识别和治疗感染。深度学习技术中的计算机视觉为寻找解决方案开辟了农业领域的途径。这项研究致力于通过应用图像处理和深度学习技术开发端到端框架,以检测椰子树中红掌象鼻虫的茎出血病,叶枯病和害虫感染。通过使用流行的分割算法来轻松定位异常边界,对一组健康和不健康的椰子树图像的手工收集图像进行了分割。定制设计的深度2D卷积神经网络(CNN)受过训练,可以预测疾病和害虫感染。同样,最先进的Keras预训练CNN模型VGG16,VGG19,InceptionV3,DenseNet201,MobileNet,Xception,通过归纳转移学习方法,对InceptionResNetV2和NASNetMobile进行了微调,以将图像分类为受感染或健康。实证研究确定,k均值聚类分割比阈值分割和分水岭分割方法更有效。此外,InceptionResNetV2和MobileNet的分类准确度分别为81.48%和82.10%,科恩的Kappa值分别为0.77和0.74。手动设计的CNN模型的Kappa值为0.91,验证精度达到96.94%。通过微型Web框架Flask在Web应用程序中部署了MobileNet模型和定制的2D-CNN模型,以自动检测椰子树病或害虫感染。通过归纳转移学习方法,对NASNetMobile和NASNetMobile进行了微调,以将图像分类为受感染还是健康。实证研究确定,k均值聚类分割比阈值分割和分水岭分割方法更有效。此外,InceptionResNetV2和MobileNet的分类准确度分别为81.48%和82.10%,科恩的Kappa值分别为0.77和0.74。手动设计的CNN模型的Kappa值为0.91,验证精度达到96.94%。通过微型Web框架Flask在Web应用程序中部署了MobileNet模型和定制的2D-CNN模型,以自动检测椰子树病或害虫感染。通过归纳转移学习方法,对NASNetMobile和NASNetMobile进行了微调,以将图像分类为受感染还是健康。实证研究确定,k均值聚类分割比阈值分割和分水岭分割方法更有效。此外,InceptionResNetV2和MobileNet的分类准确度分别为81.48%和82.10%,科恩的Kappa值分别为0.77和0.74。手动设计的CNN模型的Kappa值为0.91,验证精度达到96.94%。通过微型Web框架Flask在Web应用程序中部署了MobileNet模型和定制的2D-CNN模型,以自动检测椰子树病或害虫感染。实证研究确定,k均值聚类分割比阈值分割和分水岭分割方法更有效。此外,InceptionResNetV2和MobileNet的分类准确度分别为81.48%和82.10%,科恩的Kappa值分别为0.77和0.74。手动设计的CNN模型的Kappa值为0.91,验证精度达到96.94%。通过微型Web框架Flask在Web应用程序中部署了MobileNet模型和定制的2D-CNN模型,以自动检测椰子树病或害虫感染。实证研究确定,k均值聚类分割比阈值分割和分水岭分割方法更有效。此外,InceptionResNetV2和MobileNet的分类准确度分别为81.48%和82.10%,科恩的Kappa值分别为0.77和0.74。手动设计的CNN模型的Kappa值为0.91,验证精度达到96.94%。通过微型Web框架Flask在Web应用程序中部署了MobileNet模型和定制的2D-CNN模型,以自动检测椰子树病或害虫感染。手动设计的CNN模型的Kappa值为0.91,验证精度达到96.94%。通过微型Web框架Flask在Web应用程序中部署了MobileNet模型和定制的2D-CNN模型,以自动检测椰子树病或害虫感染。手动设计的CNN模型的Kappa值为0.91,验证精度达到96.94%。通过微型Web框架Flask在Web应用程序中部署了MobileNet模型和定制的2D-CNN模型,以自动检测椰子树病或害虫感染。

更新日期:2021-01-28
down
wechat
bug