Automatic Identification Algorithm Of Early Corn Seedlings And Weeds

With the fast expansion of the computer technology,the image processing and pattern recognition technology is growing very fast.Also,the detection problem is one of the main tasks in the field of the computer vision.In addition,the detection task is used extensively in our daily practical applications.In the domain of agricultural production,the cereal products is one of the most important sources of foods in our country.Corns are important components in the cereal products,the healthy growth of the corn seeding is very important to both the safety of the food and industrial production.So it is very necessary to remove the weeds around the corn automatic.People usually use the traditional way to remove weeds including removing the weeds manually and using the herbicide and so on.Although the traditional ways may achieve a wonderful result,it consume a lot of manpower cost and raise cost.Using the herbicide will pose a threat to the people who eat the corns.In addition,herbicides will cause the environment worse.Besides,if we use the herbicide in a long period of time,the soil will has strong dependence on the herbicide,so using the herbicide to remove weeds is also not a very good method.So automatic weeds removal is a necessary and urgent task.Based on the problems we have proposed,the computer vision and the method of deep learning provide the basis for the studies on our problem.Our target is to distinguish the corn and weeds automatically and efficiently.This paper used the method in the field of the detection problem in computer vision to propose the ways of distinguishing the corn and weeds.We make many experiments in our data set to verifying the method we have proposed.This paper firstly analyzing the shortcomings of traditional methods of removing weeds and the intelligent weeding has a very high requirement for accuracy and high performance in speed.After that,we created our data set by shooting large amounts of corn and weed pictures in the greenhouse.We observed the main difference between corn seedlings and weeds as well as using some processing methods in the field of computer vision.Firstly,we do some preprocessing to the collected data sets in order to get rid of the influence of light and noise,two methods were used to realize automatic identification of early corn seedlings and weeds.The first way is the method of manually selecting features.The features were selected by observing the main difference between early corn seedlings and weeds.After the features were extracted,we fused the two featuresand then extracting feature vectors and the classifier is used to train the traditional classification vectors.Finally a classification model of maize seedlings and weeds can be obtained.The second way is by means of depth learning.As we know,the features of convolutional neural networks can be classified,based on the current popular Faster R-CNN detection model.We used the region proposal network RPN and the Fast R-CNN detector for classification.We made annotations to our data set manually and adjust the construction and parameters of the network.We trained the network on our own data sets and finally the obtained models can be used for distinguishing the corn and weeds.We start with practical problems throughout the process,using the knowledge in the field of computer vision and proposing the methods for automatic identification of early seedling and weed.Finally,we make a summary of the work of this paper and point out the shortcomings of our work and what we want to do in the future.