Pollen Cell Image Recognition Based On Data Augmentation And Feature Fusion

Pollen cell image recognition plays an important role in environmental and medical fields such as air detection,fossil identification,honey quality control,plant dating and tracking.Traditional pollen cell image recognition is based on artificial feature design,and machine learning method is used for classification and recognition,which not only requires the plant morphology knowledge and practical experience of practitioners,but also requires complex feature engineering.The image recognition of pollen cell based on Convolutional Neural Network can avoid the uncontrollable precision error caused by complex operation,and achieve the efficient and reliable classification and recognition of pollen image.For this reason,this paper studies the image recognition of pollen cells based on Convolutional Neural Network.The main work and contributions are summarized as follows:(1)Aiming at the problem of small data set caused by the difficulty of image acquisition and imaging of pollen cell,a method of pollen cell image recognition based on data augmentation is designed.On the basis of POLEN23E pollen cell image data,three kinds of segmented pollen cell image data sets are added,and the data set of POLEN23E is statistically expanded based on Mixup algorithm.Multiple augmented pollen cell image data sets such as MixupPOLEN23E+CPOLEN23E are constructed.The experiment shows that the image set augmented by MixupPOLEN23E+CPOLEN23E has achieved good recognition effect.The data augmentation method in this paper improves the recognition accuracy of pollen cell image by an average of 3.4%-4.6%,effectively making up for the lack of sample size.(2)Aiming at the difficulty of pollen cell image recognition,a method of pollen cell image recognition based on feature fusion is designed.On the one hand,the traditional feature extraction of pollen image,the construction of visual dictionary based on word frequency,and the cascading fusion with the Deep Residual Network feature map,strengthen the feature learning of samples.On the other hand,the focal loss function is introduced into the network to reduce the difference between simple and difficult samples.Experiments show that the recognition method based on feature fusion can effectively improve the quality of feature extraction of pollen cell image and improve the recognition accuracy of pollen cell image.