Study On Jujube Yield Estimation Model Based On Image Processing

In recent years,the planting area of jujube trees has increased,and the sudden increase in production has led to an oversupply in the jujube market,and there has been a slowmoving phenomenon in many areas.Estimating the yield of jujube in advance helps the farmer to contact the processing purchaser in advance or prepare enough storage space to reduce losses.At present,the main method for measuring fruit trees in China is to estimate the time by using a certain measurement formula.The loss of time and manpower is large,and the selection of random samples is prone to bias and has limitations.With the improvement of computer performance and the improvement of data collection and storage capabilities,artificial intelligence technology has developed rapidly.Image processing technology has begun to be applied to various fields,including image processing based fruit tree yield estimation.Because jujube fruits and leaves are similar in color and fruit targets are small,fruit identification and yield estimation of jujube trees is a comprehensive problem with multiple challenges.This paper studies the estimation model of jujube yield based on image processing.The main research contents are as follows:First,build a data set and choose a deep learning framework.Since the open data set does not contain image data of jujube trees,this experiment uses Python to crawl images of this category in the network and collect related images using a digital camera,and enhancement train set.The image data is then annotated with LabelImage and all data is organized into a data set in VOC format.Then compare and analyze the commonly used deep learning framework,and choose Tensorflow framework as the learning framework of jujube fruit detection network.Second,a jujube fruit detection network based on candidate regions was constructed.In this paper,five kinds of classical feature extraction networks are compared and analyzed,and ResNet101 is used as the basic feature extraction network.Using the Faster R-CNN with high recognition accuracy as the algorithm,a deep convolutional network for jujube fruit detection is constructed.Jujube fruits were identified and the three characteristic parameters of the proportion of fruit areas required,number of fruits,and maturity for measuring production.Then,through the experiment of controlling variables,the Anchor selection strategy,learning rate and batch size are adjusted to obtain the best training effect.Third,BP neural network was selected for yield estimation.A threelayer neural network was used to construct the yield estimation model.The selected three characteristic parameters were used as the input of the neural network,and the jujube yield was used as the output to establish the BP neural network model.By adjusting the number of nodes in the hidden layer and parameters in the neural network,the parameter training network with the best test effect is selected.