The Identification Technology Research Of Corn Seedlings And Weeds Based On Machine Vision To Target Application System

The prevention and control of diseases and insect pests and weeding in corn fields in China mostly adopt the traditional uninterrupted continuous spraying operation mode,and precise application of corn seedlings or weeds cannot be carried out,which increases the cost of preventing and controlling diseases,insects and weeds and causes environmental pollution.In order to solve the above-mentioned problems,the research on the identification technology of accurate target application is carried out.The target application technology can distinguish the crop object and the soil background,and effectively reduce the use of the medicine.Aiming at the problem of low accuracy of seedling and grass identification in the complex environment of corn in the seedling stage,this paper conducts research on the recognition technology of corn seedlings and associated weeds,and uses deep learning methods to train the seedlings and grass data sets and improve the recognition of corn seedlings and weeds.With the accuracy of segmentation,the medicine is sprayed on the weed application target in a timely,quantitative and accurate manner.The main work of the paper is summarized as follows:(1)Introduced the basic network structure of convolutional neural network,discussed the target detection algorithm based on deep learning at this stage,and carried out the core principles and basic detection process of Faster R-CNN and Mask R-CNN.Explained that through the comparative analysis of the network structure and experimental results of the two algorithms,it is found that the Mask R-CNN algorithm can not only realize the segmentation and area calculation of the seedlings and grass,but also has a higher level of performance on the basis of satisfying the real-time detection of seedlings and grasses.Seedling and grass recognition accuracy,so this paper chooses to use Mask R-CNN as the basic network for corn seedling and weed recognition in this paper.(2)Constructed a weed segmentation model based on the improved Mask R-CNN algorithm.First,collect a large number of pictures of field corn seedlings and three main accompanying weeds as sample sets,and design a Python program to make the json annotation file perform data enhancement operations with the pictures at the same time,which effectively reduces the workload of manual annotation of the expanded data set.The field data set for the training of seedlings and grasses in this article;secondly,the Mask R-CNN algorithm of this article is optimized: one is to introduce the Res Ne St optimized feature extraction network that uses segmented feature map attention by channel,and the other is to use softened non-polar Soft Non-Maximum Suppression(Soft-NMS)improves the regional suggestion suppression method.Finally,use the segmentation algorithm in this paper to compare the training loss value with the feature extraction network before optimization and the non-maximum suppression method.The experimental results show that the improved Mask R-CNN has a lower total training loss value,with an average accuracy(Mean Average Precision,m AP)of 0.482.The algorithm in this paper effectively improves the accuracy of seedling and grass detection.(3)Build a test bed for target application based on Mask R-CNN seedling and grass segmentation.The test bench and the overall plan of the target application system were designed,and the corn seedling and weed positioning research was carried out.On the Ubuntu 18.04 system,combined the image acquisition module,the Mask R-CNN-based crop and weed information detection module,the target application control module and the application system status detection module to design the upper computer interface of the field corn and weeds to the target application system,set up a machine vision-based test platform for spraying to the target to carry out spraying test to the target under different light intensities and different degrees of occlusion of the seedlings.Through the research in this article,the improved Mask R-CNN algorithm can adapt to the detection of seedlings and grass targets in a complex application environment.After verification with the target application system,the algorithm has good recognition accuracy and speed,and it provides certain technical support for the improvement of crop information acquisition technology of field intelligent equipment.