Research On The Identification Method Of Pine Forest Colorful Trees Based On Remote Sensing Images

Pine wood nematode disease is a highly contagious plant epidemic with a mortality rate of up to 100%.In recent decades,it has become one of the most serious and dangerous forest diseases and pests in China.In addition to causing great damage to the ecological environment,it also brings huge losses to the economy.Therefore,in response to the severity of pine wood nematode disease,timely and accurate detection of infected trees to prevent their spread is crucial for preventing and controlling pine forest diseases and protecting pine forests.This article proposes using RGB aerial images obtained by drones as the research object,selecting the forest farm in Lishui District,Nanjing City as the research area,and using machine learning methods to identify pine wood nematodes and discolored trees.The specific content is as follows:(1)Construct a color changing wood recognition model based on support vector machine.We use directional gradient histograms and convolutional neural networks to extract features from images,and use principal component analysis to reduce the dimensionality of feature vectors.The model was trained using radial basis function,polynomial kernel function,linear kernel function,and sigmoid kernel function.The results showed that the recognition model based on convolutional neural network using radial basis function had the best performance,with a test set accuracy of 81.09%.However,the overall recognition accuracy of this method is not high,and there are errors,omissions,and inaccuracies in the recognition results,requiring the construction of a deeper network model.(2)In response to the shortcomings of traditional machine learning models in identifying pine wood nematode discolored wood,a discolored wood recognition method based on YOLOv5 is proposed.YOLOv5 achieved an average accuracy of 86.88% in identifying discolored trees infected with pine wood nematodes,with an accuracy rate of 87.46%.However,the duration of infection of pine wood nematode infected trees varies,resulting in varying degrees of discoloration,which increases the difficulty and misjudgment rate of the model in identifying pine wood nematode infected trees.(3)A recognition algorithm based on improved YOLOv5 is proposed to address the issues in the above recognition.Firstly,a new prediction header constructed by Transformer is introduced to reduce computational complexity;Secondly,introduce CBAM attention mechanism to reduce the interference of background information;Finally,add YOLOv8’s C2 F module to ensure rich gradient flow information.The results showed that the improved YOLOv5 model successfully reduced the missed detection rate in the experiment,improved the recognition rate for covered targets,and increased the average accuracy AP by 3.92%,with an accuracy increase of 5.38%.(4)The results of three recognition models based on support vector machine,YOLOv5,and improved YOLOv5 were compared.The recognition model based on YOLOv5 improved the average accuracy of the recognition results by at least 11.6% compared to the support vector machine,while the improved YOLOv5 algorithm improved the average recognition accuracy by 3.92% and the accuracy by 5.38% compared to the original...