Weed management using remote sensing
- 1. Professor Jayashankar Telangana State Agricultural University College of Agriculture, Rajendranagar, Hyderabad AGRON 503 SUBMITTED BY M. VEERENDRA RAM/18-07 DEPARTMENT OF AGRONOMY Submitted to Dr.M.Madhavi Principle Scientist & Head AICRP- Weed Management
- 2. WEED MANAGEMENT USING REMOTE SENSING
- 3. Introduction • The current weed management practise includes spraying the whole agricultural field with chemical herbicides. • Although this seems to be effective, it has huge effect on the surrounding environment due to excessive use of chemicals. • As weeds don’t spread over the entire field it is waste to spray herbicide to the entire field. • Precision agriculture is a farming management concept based on observing measuring and responding to the variability in the crops.
- 4. Precision agriculture/ Satellite farming/Site specific crop management is the application of technologies and principles to manage spatial and temporal variability associated with all the aspects of agriculture to improve the productivity.
- 5. Remote sensing - sensing things from a distance. • Art and science of obtaining useful information about an object with out being in physical contact with it without physically contact between the object and sensor. • Remote sensing uses the electromagnetic spectrum to image the land, ocean and atmosphere. All objects on the surface of the earth have their own characteristic spectral signature
- 6. Figure 1. The process of Remote Sensing
- 8. Unmanned aerial vehicles Air based platforms
- 9. Unmanned ground vehicles Ground based platforms
- 10. Recently, applications of remote sensing using UAVs have shown great promise in precision agriculture as they can be equipped with various imaging sensors to collect high spatial, spectral, and temporal resolution imagery
- 13. Figure 2. reflectance and absorbance of different wavelengths by different components
- 14. Figure 3. spectral reflectance of different weeds Mustafa et al., 2013 Grass lands of Texas, USA1)
- 15. Figure 4.The difference in reflectance in a weedy plot and weed free plot is due to more green vegetation covering the soil. Chang et al., 2009 2)
- 16. One of the important and challenging components of SSWM is weed recognition and field mapping for an appropriate early automatic weed control (Shaner and Beckie,(2014). However, the reflectance characteristics of crops and weeds are generally similar in their early growth stages, thus imposing additional difficulties to discriminate between them (López-Granados,2011; Perez-Ortiz et al., 2016). Moreover, weeds can grow in small patches in the early season, which also adds challenges and requires high resolution imagery to detect them.
- 18. Figure 5.Example of the spatial and spectral combination results using an SVM classifier. (a)Multispectral orthoimage; (b) Crop (green) and weed (red) location deduced from spatial information; (c) Weed (green) and crop (red) location deduced from spectral information; (d) Weed(green) and crop (red) location deduced from the combination of spatial and spectral information. Marine et al., 2018 France, Maize crop having weeds like Chenopodium album, Cirsium arvense3)
- 19. A study was conducted at the experimental fields of the Institute for Agricultural and Fisheries Research (ILVO) in the agricultural region of Merelbeke, which is located in East- Flanders Province, Belgium. The area of the maize plot was about 150m2. The naturally infested weed species in the maize plot included Convolvulus arvensis, Chenopodium album and Digitaria sanguinalis species Junfeng et al.,2017 4)
- 20. Figure 6.Partial view of image processing steps for inter-row weed detection. (a) raw RGB image, (b) vegetation index of ExG, the dark pixel represents background and the light pixel represents vegetation, (c) binarized image via Otsu’s algorithm, (d) pre-processed binary image, (e) Canny binary edge image, (f) detecting maize row line by the Hough algorithm, (g)masking maize crop, (h) inter-row weed binary image
- 21. Figure 7. Intra-row weed classification result from the OBIA. (a) The classification results by the RF (b) large view from marked yellow box area. • Concerning UAV-based remote sensing, Object-based Image Analysis (OBIA) is a common methodology in classifying objects. The OBIA first identifies spectrally and spatially homogenous objects according to its segmentation results and then it combines spectral, textural and geometry information from objects to boost classification results
- 22. OBIA
- 23. Figure 8.Procedure steps of fusion of intra- and inter row weed. (a) the partial view of intra-row weed binary image, (b) the partial view of inter- row weed binary image, (c) the partial view of inter and intra row weed fusion result, (d) the detected weeds marked with the red bounding box. Junfeng et al.,2018
- 24. OBIA Pixel based Figure 9.Procedure to distinguish crop and weeds( inter row and intra row)
- 25. 11(a) The OBIA weed predictions for the 20 selected windows, (b) the relationship between the OBIA predictions and the GT of weed densities. 10(a) The Hough weed predictions for the 20 selected windows, (b) the relationship between the Hough predictions and the GT of weed densities
- 26. 12(a) The weed predictions from the fusion of the Hough features and OBIA for the 20 selected windows, (b) the relationship between the values from the proposed method and the GT of weed densities. Junfeng et al.,2017
- 27. • Sugarcane is a long duration crop which reaches its maturity in 11–12 months. Crop growth is very slow at the initial stage i.e. it takes 25–30 days to complete germination and another 90–95 days to complete tillering • Weeds in these sugarcane fields are classified as grasses, sedges, broad leaved weeds and climbers wherein Cynodon dactylon, Panicum species, Sorghum halopense, Chloris barbata, Dactyloctenium aegyptium are family of grasses, Cyperus iria and Cyperus rotundus are family of sedges, Trianthema portulacastrum, Amaranthus viridis, Portulaca oleraceae, Commelina bengalensis, Cleome viscosa and Chenapodium album are broad leaved weeds and Convolvulus arvensis, Ipomea sepiaria and Ipomea alba are climbers 5)
- 28. • The weed detection system has four major steps. They are: • (i) Colour based greenness identification • (ii) Texture extraction • (iii) Feature vector generation • (iv) Classification. • The system also considers the surface texture of the leaf parts (venation) rather than the size and shape of the individual leaf Fuzzy real time classifier
- 29. Figure 13.Data set regarding the crop
- 30. Leaf-texture and venation extraction from 10 species • a.Ipomea alba • b.Convolvulus arvinse • c.Coccinia grandis • d.Trianthema portulacastrum • e.Sugarcane • f.Amaranthus viridis • g.Cyanotis axillaris • h.Physalis minima • i.Comalina bengalensis • j.Cyperus rotundus. Figure 14.Data set regarding the weeds
- 31. Colour based navigation process weed detecting vehicle
- 32. Figure 15.Architecture of the weeding robotic model.
- 33. A field robotic model in which the weed detection algorithm is implemented has been tested in different sugarcane fields and it gives an overall accuracy of 92.9% and processing time of 0.02 s. Sujaritha et al., 2017
- 34. • Monitoring high density of plants (mache salad) with the undesired presence of weeds is challenging since the intensity or color contrast between weed and crop is very weak. • Therefore, this problem is well adapted to test scatter transform which is a texture-based technique. • A scattering transform builds a stable informative signal represents for classification. It is effective in image, audio and texture discrimination. France, Mache salads Pejman et al., 2019 6)
- 35. Figure 16. view of the imaging system fixed on a robot moving above mache salads of high density with weeds
- 36. RGB images from top view for the detection of weed out of plant used as testing data-set
- 37. Simulation pipeline for the creation of images of plant with weed Figure 17.Illustration of different types of weeds used for the experiment
- 38. Figure 18.Anatomical scales where (Wi,Pi) presents the scales of weeds and plants respectively; (W1, P1) points toward the texture of the limb, (W2, P2) indicates the typical size of leaflet and (W3, P3) stands for the width of the veins. Sw and Sp show the size of a leaf of weed and plant, respectively. The classification of weed and plant is done at the scale of a patch taken as 2 max(Sp, Sw) in agreement with a Shannon-like criteria
- 39. Figure 19. Output images for each class (weed on left and plant on right) and for each layer m of the scatter transform. In the application of scatter transform to classification found in the literature so far, the optimization of the architecture was done.
- 40. 20(a) (97.27%) accuracy 20 (b) (69.45%)accuracy Visual comparison of the best and the worst recognition of weeds and plants by scatter transform. Pejman et al., 2019
- 41. Figure 21.Visualization on Drop-on-Demand herbicide application Norway, Carrot field having weeds like Chenopodium album, Poa annua, Stellaria media Utstumo., 2019 7)
- 42. The 2017 Asterix robot prototype in field trials in Central Norway.
- 43. Figure 22.Two uppermost images: same plot (plot No. 1001) before and 17 days after glyphosate application with the robot. Two bottom images: untreated plot (plot No. 1002)
- 44. • The label application for Glyfonova Plus ranges from 540 g/ha to 2880 g/ha depending on the types of weeds and weed pressure (Cheminova AS, 2015). • A treatment scheme with the robot and the DoD system, would consist of 2-3 treatments in combination with mechanical weed control in between the rows. Building on the experience from the lab and field trials, we would estimate a total application of 50 - 150 g/ha glyphosate. This represent a ten-fold reduction in applied herbicide. Utstumo., 2019
- 45. References: . Dwivedi, A., Naresh, R.K., Kumar, R., Yadav, R.S. and Rakesh,K.,2017:Precision agriculture. Gaoa, J., Liaob,W., Nuyttensc,D., Lootensd,P., Vangeytec,J., Pižuricab,A., Hee,Y. and Jan G. Pietersa,2017:Fusionofpixelandobject-based featuresforweedmappingusingunmanned aerial vehicle imagery. International Journal of Applied Earth Observation and Geoinformation, 67: 43-53 Hassanein,M. and El-Sheimy,N. An efficient weed detection procedure using UAV imagery system for precision agricultural applications. Remote Sensing and Spatial Information Science. Ishaq ,W.I.W., Hudzar,R.M. anRidzuan,R.M.N. 2011:Development of variable rate sprayer for oil palm plantation. Bulletin of polish academy of sciences ,59:3 Lottes,P., Behley,N., Milioto,A. and Stachniss, C. 2018:Fully Convolutional Networks with Sequential Information for Robust Crop and Weed Detection in Precision Farming. IEEE Robotics and Automation Letters Peña,J.M., Sánchez,J.T. and Pérez, A.S, 2015: Quantifying Efficacy and Limits of Unmanned Aerial Vehicle (UAV) Technology for Weed Seedling Detection as Affected by Sensor Resolution. Sensors , 15: 5609-5626 Rasti ,P., Ahmad, A., Samiei ,S., Belin,E. and Rousseau,D.2019:Supervised Image Classification by Scattering Transform with Application to Weed Detection in Culture Crops of High Density. Remote Sensing, 11: 249 Tamouridou ,A.A., Alexandridis, T.K. , Pantazi ,X.E. , Lagopodi,A.L., Kashefi ,J., Kasampalis ,D., Kontouris ,G and Moshou,M.2017:Application of Multilayer Perceptron with Automatic Relevance Determination on Weed Mapping Using UAV Multispectral Imagery. Sensors , 17: 2307 Utstumoa,T., Urdala,F.,Brevika,A.,Døruma,J.,Netlandc,J. and Overskeida. 2019:Robotic in-row weed control in vegetables, Computers and Electronics in Agriculture Xu,Y., Gao,Z., Khot,L., Meng,X and Zhang ,Q .2018:A Real-Time Weed Mapping and Precision Herbicide Spraying System for Row Crops.Sensors,18: 4245 Zisi ,T., Alexandridis ,T.K., Kaplanis,S., Navrozidis ,L., Tamouridou,A.A., Lagopodi ,A., Moshou,D. and Polychronos,V. 2018: Incorporating Surface Elevation Information in UAV Multispectral Images for Mapping Weed Patches.Journal imaging, 4: 132