Research On Plant Image Recognition With Complex Background Based On Convolution Neural Network

The digital management and service of plant image are indispensable for the study of plant species and characteristics,the protection of plant diversity,and the guidance of agricultural practitioners in the scientific and reasonable use of plant production.Therefore,the accurate recognition of plant image is particularly important.In recent years,most of the plant image recognition is limited to the single background leaf and flower images,and the recognition method is to set and select the features manually,then encode the features,and finally combine with the classifier for classification recognition.However,most of the plant images collected in our daily life are plant images under complex background,which are using the traditional image In image recognition,it is difficult to design and extract features,and the recognition rate cannot be guaranteed.To solve these problems,this thesis focuses on the plant image recognition method based on convolution neural network in complex background.This thesis focuses on two aspects: on the one hand,it focuses on the analysis of the recognition effect of plant image under the convolution neural network model of AlexNet and Google Net.Using the convolution neural network technology,features are automatically learned from big data to avoid the limitations of artificial setting and selection of features,so as to ensure the recognition rate of plant image under complex background to a certain extent.On the other hand,in order to improve the accuracy of complex background plant image recognition,considering the background interference in complex background plant image recognition,based on the classic convolution neural network,this thesis improves the plant image recognition method,from the perspective of effective and invalid features of the image,from the aspect of target segmentation,an effective region extraction and selection mod el based on Mask R-CNN is designed,and a convolutional neural network plant i mage recognition method based on effective region extraction and selection is proposed.This method first trains an effective region extraction model based on Mask R-CNN using labeled MLT data set,the purpose of this model is to segment the effective regi on in the picture,and then train an effective region selection model through convolution neural network(CNN)using the image(flower,leaf)data set.The purpose of this model is to make the data set retain the effective areas such as flowers and leaves,which can represent the plant image categories,and to discard the areas where the features are not obvious,so as to ensure that the extracted image features are more accurate and thus reduce the complexity of image recognition,a plant image recognition model MRC-GoogleNet based on Google Net is designed in combination with the effective region extraction and filtering module.From the point of view of complex background and effective region selection,the model breaks the limitation that plant image recognition is limited to single background leaves and flowers,and effectively improves the adaptability of recognition algorithm;in addition,the sample data used for training in the recognition model in this thesis is selected from the original image,so the features extracted from the image recognition model can highlight the characteristics of plant categories,which not only ensures that the effective features of plant images are not missed,but also avoids the interference of invalid features.Experimental results and data show that MRC-Google Net model can extract more effective image features and improve the recognition accuracy compared with the classical convolutional neural network model.