Research On The Algorithm For Intelligently Identifying The Growth Status Of Corn

As one of the three major cereals in the world,corn is not only rich in sufficient dietary fiber,lutein and other nutrients,but also has the effects of clearing away dampness and heat,delaying aging,and benefiting liver and gall,which is very important for China’s agricultural development.However,in the field of corn planting,large-scale planting of corn requires timely control of corn growth status,so timely and effective monitoring of corn growth status is very important.The traditional monitoring of corn growth status mainly depends on two methods,one is manual monitoring,but this method is empirical judgment,which is not only complex,error prone,subjectively affected and inefficient,and cannot meet the demand for accurate judgment of corn growth status when large-scale cultivation of corn;The other method is an image recognition algorithm based on computer vision,but the feature quantity needs to be manually extracted,which greatly tests the experience accumulation of engineers themselves.Therefore,the extracted features are affected by human factors,which is easy to cause information loss,thus having an unexpected impact on the accuracy of corn growth state recognition.Recently,with deep learning widely applied in various fields,its excellent detection effect has also been widely recognized.Therefore,this paper designs an intelligent recognition algorithm of corn growth state based on the improved Faster RCNN(Faster region convolution neural network),which is committed to improving the detection accuracy of corn tassels,and determines whether the corn in the field enters the heading stage by detecting the proportion of corn tassels,so as to replace manual monitoring.First of all,we propose to improve the feature extraction module based on Res Net residual network.In view of the complexity of the field environment in which corn plants grow,in order to make Faster RCNN network perform better on the detection and recognition of corn tassels and reduce the impact of other environmental factors such as shadows and light,this paper plans to use Res Net101 to replace the VGG16 used in the original Faster RCNN network as the feature extraction network,aiming to improve the average accuracy of the entire detection model.Secondly,it is proposed to redesign the size of anchors to improve the Regional Proposal Network(RPN).In view of the low matching degree between the anchor’s corresponding size proportion and the maize tassels in the module and the inaccurate prediction of the candidate box,this paper plans to redesign the anchor’s size and proportion during the generation of the candidate box,so as to better cover the characteristic area of the maize tassels.Finally,NMS(non maximum suppression)algorithm is proposed to improve the non maximum suppression algorithm in RPN module.When screening candidate boxes,in order to further filter the best candidate box,this paper plans to use Soft NMS to replace the NMS algorithm,so as to avoid the problem of missed detection when the detection targets overlap,so as to further improve the average detection accuracy of the entire network.In this paper,the algorithm of identifying corn growth state based on convolutional neural network is studied.Compared with traditional methods,it not only improves the average accuracy of detection,but also avoids the error of manual participation in extracting feature quantities.According to the characteristics of maize tassels,the Faster RCNN network is improved to improve the average detection accuracy of the whole network,which provides a theoretical basis and reference value for large-scale planting of maize in the actual production process.