Study On The Identification Of Condition Potential Of Rice Growth Based On Convolution Neural Network

In order to respond positively to the notification issued by the Ministry of Agriculture on the case of agricultural and rural big data practice,this paper takes rice growth stage images as the starting point for research.The growth status of rice in each period directly affects the yield and quality of rice.Therefore,judging the normal rice and abnormal rice in the growth stage of rice in a timely manner has become the top priority of agricultural development.The traditional manual observation method is inefficient and consumes a large amount of human resources.Its observational results are subjective;the use of remote sensing monitoring is susceptible to cloud-rain weather and has a wide range of observations,and is not suitable for quickly and accurately rice in small-scale paddy fields.Growing up to monitor and determine.In order to make up for the deficiencies of the above methods,this study used the convolutional neural network with unique advantages in the field of machine vision to identify two-dimensional images to study and analyze the images of the growth stages of the rice harvested by fixed-point cameras in agricultural parks.It is expected that the degree of automatic and rapid discrimination of the growth and development stages of the rice growth stage will be enhanced through deep learning of the machine.Frist,collect samples and establish a database.The data from this experiment was collected from experimental rice fields in agricultural science and technology parks such as Duerbote Mongolian Autonomous County of Daqing City in Heilongjiang Province and Qihehalehe City in the Heilongjiang Province;the reference standards for the classification of normal rice and abnormal rice in the growth stage of rice were constructed by consulting the relevant literature.The discriminant basis is provided;this article classifies the images of rice growth stage into 6 categories.The K nearest neighbor classifier method is used to classify the images of each type of training images and test images.Next,construct a Convolutional Neural Network Model for the Growth Stages of Rice Growth.After the original image is subjected to preprocessing operations such as shear transformation and gray threshold segmentation,and a network model of input-convolution-pooling-convolution-pooling-full-connection-full-connection-output is established.The nucleation size is 5*5,the pooling layer adopts max-pooling method,and the output adopts Softmax classifier.240 small batch samples and 3600 large batch samples are tested through convolution and sampling operations respectively,and 200 iterations of the model are used.The accuracy rate was 87.0588% and 93.5936%.Small batch sample tests verify the reliability of the model,and large-scale sample tests verify the generalization ability of the model.Last,The particle swarm optimization(PSO)algorithm was used to optimize the convolutional neural network model of rice growth and development.By backpropagation optimization of the weights between the layers in the network,the particle search global optimal solution to optimize the growth and development stage of rice growth model,the model has a faster convergence rate and higher accuracy.The network weights are assigned to the particles by the encoding operation,the particles are given the network weights by the decoding operation,and parameters such as the number of populations,the inertia,and the learning factor variables are set.After the optimization,240 small batch tensor samples and 3600 large batch samples were tested.The accuracy rate of the model was 97.2633% and 99.0133%.The optimized model has good performance and stability.For the above studies,this paper provides a novel and scientific effective method for identifying the superiority and inferiority of rice.At the same time,it also expands the application of convolutional neural networks in the field of agricultural crop growth,providing a theoretical basis for subsequent research.