Research On Automatic Recognition Of Corn Development Period Based On Deep Learning

Obtaining information on the growth and development of crops accurately can carry out fertilization and irrigation scientifically,and promote the development of modern agriculture effectively.Corn is one of the main crops in China.The environment and climate of the planting area are complex and changeable.It is an important task to obtain the growth and development status and carry out related agricultural operations in time.Therefore,research on the automatic identification method of corn development is a research hotspot in the field of crop growth observation.In order to solve the problem that the traditional automatic recognition method of corn development has low feature extraction efficiency and poor recognition effect,this paper introduced deep learning into the automatic identification research of corn development period,proposed a corn development period identification method based on convolutional neural network,combined with transfer learning,image average segmentation preprocessing and target detection algorithm based on multi-dimensional feature map,established and improved the convolutional neural network model,achieved efficient corn development period automatic identification.The specific research contents are as follows:The basic structures of convolutional neural network were studied.The standard convolution and lightweight convolution methods were analyzed.The standard convolutional neural network framework and the lightweight convolutional neural network framework were constructed respectively.The samples in corn development training dataset were used to train the convolutional neural networks.The standard convolutional neural network model and the lightweight convolutional neural network model were obtained to predict the corn development period test samples.The results show that compared with the traditional method,the prediction accuracy of the convolutional neural network model is higher,and the training of lightweight convolutional neural network recognition model is faster than the training of standard convolutional neural network recognition model.Aiming at the problem of over-fitting of lightweight convolutional neural network recognition model caused by insufficient sample,this paper used the transferring parameter method of transfer learning,transferred the model parameters trained by large dataset ImageNet to convolutional neural networks in different proportions,selected the best recognition model according to the test dataset recognition rate,and compared the test dataset recognition rate with training dataset recognition rate.The results show that the transferring parameter method can effectively improve the lightweight convolutional neural network recognition model’s generalization capabilities.Aiming at the problem of low recognition rate of the three-leaf period and the tasseling period,it is found that the three-leaf period and the tasseling period images lose many important features in the pre-processing process by analyzing the training set samples.This paper used the average segmentation image method to pre-process the training dataset,combined the voting method with the lightweight convolutional neural network model to predict the corn developmental period samples in the test dataset.The recognition rates of the three-leaf period and the tasseling period are 84.62% and 92.41% respectively.Due to the tasseling period is easily misjudged as the jointing period,the target detection method based on the multi-size feature map is adopted,and the tassel detection model is established.The tassel detection was performed again on the image that has been determined to be the jointing period.The recognition rate of the light convolutional neural network model has been increased to 96.06%.Through the research on the recognition method of corn development period,the optimal corn development period recognition model based on convolutional neural network was established,which effectively improved the recognition rate of corn development period,and which is of great significance and application value for the development of crop automatic observation system.