Classification Of Rice And Weeds Based On UAV Remote Sensing Images

Weeds have a serious impact on rice yield.At present,the control of rice weeds mainly depends on herbicides,and most farmers use uniform spraying to control weeds.Excessive use of herbicides not only affects the quality of rice,but also causes environmental pollution.China has put forward the strategic goal of "two reductions",which aims to improve efficiency,reduce costs and protect the environment by reducing the application of pesticides and chemical fertilizers.Accurate application of pesticides based on agricultural condition analysis is one of the effective means to achieve this goal.Under the application background of weed management in paddy fields,accurate application of pesticides requires that the distribution information of weeds in paddy fields be obtained quickly.However,the traditional method based on Artificial Farming investigation has a large workload and low efficiency.Increasingly developed and improved remote sensing monitoring technology provides the possibility for rapid agricultural condition acquisition.In this paper,UAV is used to collect low-altitude remote sensing images of paddy fields,and weed recognition methods based on image processing and machine learning are studied.In order to analyze the distribution information of weeds in paddy fields and provide decision-making basis for precise spraying of herbicides by plant protection machinery in the future.The main work of this paper is as follows:（1）Using UAV to collect visible remote sensing images of paddy fields,the original data were pre-processed and calibrated into rice,weeds and others.The data were cut into small samples and divided into training set and test set.The feature extraction of the three types of data sets was carried out respectively.The color features included the maximum,minimum and mean values of the three RGB channels;and the texture features included lbpm and lbpm＿local,then the extracted features are fused and processed by PCA dimensionality reduction to form feature vectors of small samples.（2）Four machine learning algorithms,KNN,SVM,BP neural network and Ada Boost,are used to classify image samples,and the generalization performance of the algorithm is tested on the test set.The results show that KNN classification accuracy is 87.61%,SVM classification accuracy is 90.67%,BP neural network classification accuracy is 86.81%,Ada Boost classification accuracy is 89.22%.（3）In order to improve the classification accuracy,a single machine learning algorithm is combined with the idea of ensemble learning.The experimental results show that the ensemble model can improve the classification accuracy.The recognition accuracy of combination classifier and superposition classifier based on SVM and Ada Boost reached 91.21% and 91.14% respectively.（4）The convolution neural network based on deep learning is used to classify and recognize UAV images.Based on transfer learning,inception＿v3 model is used as the pre-training model,and fine-tuned on the training set in this paper.The experimental results show that the accuracy of the model on the test set is 94.80%.The results show that the classification accuracy of the feature vectors extracted in this paper can reach 90.67% by using traditional machine learning method,91.21% by optimizing the ensemble learning method,94.80% by using convolutional neural network to identify rice weeds,and automatic feature mining can simplify the steps of artificial feature extraction,and the adaptability of network structure.It has better generalization ability for different types of weeds and different growth stages.The results of this study show that the combination of UAV remote sensing and machine learning can accurately analyze the distribution information of weeds in the field,and can be used as the basis for decision-making of herbicide spraying in plant protection machinery.