Detection Of Nitrogen Content In Wheat Leaves Based On Deep Learning And Mobile Terminal Application Research

Wheat is one of the most important rations in China,and also one of the most important trade rations in China,which brings abundant benefits to China’s economy.As a result,we have higher requirements for wheat planting yield and quality standards.Nitrogen is an essential element in the process of crop growth,and it is of great significance to control the amount of nitrogen application in wheat growth period for the formation of wheat yield and quality.The traditional method can not be used for nondestructive and real-time detection because of its strong destructiveness and high equipment requirements,such as physical and chemical detection.In recent years,deep learning which as a branch of machine learning has become a research hotspot in the field of computer vision.In particular,as one of the main research algorithms of deep learning,convolutional neural network has made breakthrough progress in target detection,classification and prediction tasks.This paper takes wheat leaves in the developing stage as the object,studies the detection method of nitrogen content in wheat leaves based on CNN algorithm,and develops the Android detection system of nitrogen content in wheat leaves.The results showed that:(1)Network model training requires the support of a large data set,and too small data set size has considerable negative impact on the generalization ability of model training results.In this paper,the diversity of training samples was increased by means of data enlargement.The wheat leaf image data were preprocessed by rotation,horizontal,vertical and diagonal mirror,brightness and contrast adjustment,etc.Finally,the data set of 23,000 indoor and outdoor images was obtained by means of optimization.(2)A regression prediction model was built based on the convolutional neural network.The model construction in this paper followed the idea of Lenet-5 model,including the input layer;4 convolution layers(Conv1,Conv2,Conv3,Conv4);2 pooling layers(POOL1,POOL2)and 2 fully connected layers.In order to make the output one-dimensional,Flatten layer was used to process the input and output the predicted value of leaf nitrogen content.Through experimental comparison,the network structure constructed in this paper has the best nitrogen detection effect when the training set accounts for 80% of the sum of the training set and the verification set,the learning rate is 0.001,and the size of the convolution kernel is 3*3.(3)In order to better apply the training results of wheat leaf nitrogen content detection based on CNN algorithm into practice,this paper designed and implemented the Android detection system;Based on the preliminary market research and the analysis and comparison of agricultural application software,the function and interface were designed,and the network model obtained from training was transferred to the Android platform for practical application.Finally,50 image data of test samples were selected to test the system.The accuracy rate of indoor sample detection was nearly 83%,and that of outdoor sample detection was nearly 30%.In conclusion,the deep learning network model provides a new efficient and low consumption method for wheat leaf nitrogen content detection,which has a positive impact on the application and promotion of agricultural informatization in the field.