Weed Identification Method In Wheat Field Based On Image Processing

Wheat is one of the most common crops in China.With the improvement of people's quality of life,the planting area of wheat is also expanding.However,large-scale planting of wheat is easy to cause weeds to grow wantonly.The growth of weeds will also restrict the yield and quality of wheat,which will harm the growth of wheat.The existing weeding methods mainly rely on human eye observation,and then large-scale spraying of pesticides,which will cause great pollution to the growth environment of wheat.In view of this situation,taking image processing as the core,this paper proposes a method of weed identification in wheat field based on image processing,which realizes the fixed-point spraying of weeds,and achieves the requirements of precision agriculture and the purpose of environmental protection.The main research contents are as follows:Firstly,image acquisition and preprocessing are realized.Image acquisition mainly includes acquisition time,shooting angle and shooting height.In this paper,gamma correction,grayscale processing and noise reduction of sample images are realized by image preprocessing technology.Finally,Otsu,K-means clustering algorithm and FCM are used to segment the target and background of wheat field weed image.Secondly,the feature extraction of image is realized.Effective features can improve the classification accuracy of target samples.In this paper,we extract the shallow features of the preprocessed image,and use the feature extraction operator to extract the color space features,geometric features and texture features,and then analyze the attributes of these three types of features.Finally,Realize the classification and recognition of weed image.In this paper,SVM and MPF net are used to classify and identify weeds.In order to verify the feasibility of the algorithm,the algorithm is compared with the existing lightweight Squeezenet,Mobilenet and Shufflenet networks.The experimental results show that the accuracy rate of the proposed algorithm is 97.82%,which is improved to a certain extent compared with the traditional algorithm in detection accuracy and running speed.