Weed Identification Method And Application Based On The Near Earth Remote Sensing Technology In Corn Field

Pesticide over-spraying,causing a series of serious problems to the soil and ecology,so the study of field weed removal method is particularly important.In the field of weed identification,how to quickly and accurately access to farmland weed information in real time,as weed precision control technology is facing the primary problem.Remote sensing technology has the characteristics of access to information is sufficient: its band includes visible light,ultraviolet,infrared and many other information.But also has a means of detection,non-destructive detection of good,real-time and other characteristics can be used as an important means of field weed identification.In this paper,the corn experiment field of Shandong Agricultural University was selected,and the data of indoor corn and weed were collected with ASD FieldSpec.The time to collect data was in the period of 4-7 leaf stage weeds in July,and the weeds were growing vigorously to corn field of the greater harm,Easier to identify and distinguish.Three typical weeds were selected,namely purslane,wild amaranth and Cyperus rotundus.Twenty groups of data were collected for each crop of corn and weed.Each group of data was set to 10 samples.There were 1,200 spectral data.Through the pretreatment and analysis of spectral data,734 nm,954nm,1324 nm,1869nm four sensitive bands were selected by using the selected sensitive bands.The experimental results showed that the model accuracy was 85.8%,and the distinction between weeds and corn crops was realized to a certain extent.Among them,the recognition accuracy of corn reached 90%,and the weed wild amaranth and Cyperus rotundus showed a high degree of mixing.The classification accuracy of wild amaranth was only 63.3%,of which 30% of wild amaranth was misplaced in fragrant,the lowest weed.While using the ADC multi-spectral camera to collect outdoor data,and to a certain extent to maintain the original growth of crops and weeds.Multispectral data were used to collect 25 sets of photos for each vegetation.A series of pretreatments were carried out,such as removal of background values,extraction of morphological and texture features,analysis of three principal components by principal component analysis(PCA).Three principal components were obtained by PCA and Judgment by SVM algorithm,the final discrimination accuracy of 88%.SVM algorithm provides a feasible algorithm for the identification of weeds in the field,while taking advantage of the short time and highrecognition accuracy.The results of multispectral image recognition showed that the recognition accuracy of corn was 88%,corn with wild amaranth recognition rate lowest72%,of which 24% is mixed with sweet attached,4% mixed with corn.The accuracy of weed identification based on spectroscopy was 85.8%,and the accuracy of weed identification based on multi-spectral images was 88%.It was proved that the weeds in maize could be successfully identified based on near-ground remote sensing.The research algorithm is simple,the reaction of the optical sensor is agile,the structure is simple,the operation is convenient,and there are some advantages in terms of real-time and economy.It is a feasible method and has certain application value and reference significance.The results show that the precision of two methods of research were higher.The first kind of research method of high precision reason has two aspects: on the one hand is due to the indoor experiment,reducing the influence of the external environment,and have certain effect to reduce the error.On the other hand is the difference between the plant themselves,and we know the horse tooth and wild amaranth were dicotyledonous plants,corn and sweet as monocotyledon plants,and the leaf structure of dicotyledonous and monocotyledonous plants is different,so it is easier to distinguish.The second method of high precision is due to: corn and three different kinds of weeds shape and texture exist obvious differences,can be very good from the image technology to distinguish between corn and weeds.The ultimate aim of the weed identification method is to accurately and quickly remove the weeds in the field.In this study,it is proved that the hyperspectral and multispectral have certain application value in the weed identification direction.At the same time,taking into account the current price of hyperspectral equipment and the use of conditions are complex,so the hyperspectral applications in the weed identification will take a long time.Compared with the two methods,it is considered that the weed recognition method based on image has the advantages of simple response,easy structure and convenient operation.It has a certain advantage in terms of real-time and economy.It is a feasible method,Feature recognition is more advantageous.