Research On Corn Growth Stage Identification Based On Depth Feature Learning And Multi-level SVM

Automatic crop identification is one of the core parts of precision agriculture support technology.The growing period of traditional crops is recorded by manual observation,which has the problems of time-consuming and laborious,low efficiency,strong subjectivity,different observation standards,and difficulty in ensuring measurement accuracy.At present,image processing technology is mainly used to classify and identify crop growth periods.Since crop images are taken in the field,fixed shooting equipment is required and shooting is performed at the same distance,which requires high requirements for light and shooting angles.With the extensive application of deep learning in the field of computer vision in recent years,this thesis takes corn images as the research object,and adopts deep learning technology to identify the different growth period images of corn photographed in the field.This thesis preprocesses the corn images taken in the field,uses the convolutional neural network to extract the characteristics of the corn images,combines the PSO optimization algorithm to construct a multi-level SVM based on the binary tree structure,and realizes the classification and identification of the corn growth period.The main work is as follows:(1)Preprocessing corn images taken in the field.Combining the color characteristics of corn plants,the corn images taken in the field were preprocessed with mathematical morphological algorithms and foreground object segmentation algorithms to filter out noises such as soil,weeds,and light in the images,and to prepare for subsequently extracting the characteristics of the corn images.(2)Feature extraction of corn images using convolutional neural network.When training with convolutional neural networks,there are not enough samples and network parameters are not fully trained,which results in poor classification.Firstly,the data set is extended by the data enhancement technology such as turning,cropping and rotation.then using the pre-training model of UCI data set to preserve the trained network model.The pre-training corn image is replaced by the trained model and the model is adjusted to extract the characteristics of the corn image.(3)The particle swarm optimization algorithm is used to optimize the SVM parameters and a multi-level SVM model based on the binary tree structure is constructed.For the characteristics of the extracted corn images,the SVM model was trained with the best parameters,and the classification accuracy of the corn growth period was returned.The results of the classification experiment before and after optimization were compared to verify the validity of the method.