Docstoc

Katholieke Universiteit Leuven

Document Sample
Katholieke Universiteit Leuven Powered By Docstoc
					Katholieke Universiteit




              Spectral Weed Detection
                and Precise Spraying
          Laboratory of AgroMachinery and Processing
             Els Vrindts, Dimitrios Moshou, Jan Reumers
              Herman Ramon, Josse De Baerdemaeker

                         Research sponsored by IWT and
                the Belgian Ministry of Small Trade and Agriculture
                          Overview
     Spectral measurements of crops and
     weeds
         in laboratory
         in field
     Processing of spectral data with neural
     networks
     Precise spraying


Katholieke Universiteit Leuven
        Optical detection of weeds
     Techniques
           red/NIR detectors (vegetation index)
           image processing (color, texture, shape)
           remote sensing of weed patches
           reflection in visible & NIR light
      different detection possibilities, different scales
     Requirements for on-line weed detection:
           fast & accurate weed detection
           synchronized with treatment


Katholieke Universiteit Leuven
          Spectral weed detection
  Factors affecting spectral plant signals
    leaf reflection, dependent on species
    and environment, stress, disease
    canopy & measurement geometry
    light conditions
    detector sensitivity



Katholieke Universiteit Leuven
                                           Laboratory measurements


 Spectral analysis of plant leaves
           in laboratory
                   integrating sphere



          sample



                     spectrophotometer   computer



        Diffuse Reflectance Spectroscopy of Crop and
        Weed Leaves


Katholieke Universiteit Leuven
                                 Laboratory measurements



        Diffuse Reflectance of a Leaf




Katholieke Universiteit Leuven
                                 Laboratory measurements


                    Spectral Dataset




Katholieke Universiteit Leuven
                                 Laboratory measurements


Reflectance of crop and weed leaves




Katholieke Universiteit Leuven
                                                  Laboratory measurements


                  Spectral analysis
     stepwise selection of discriminant wavelengths
     multivariate discriminant analysis, based on
     reflectance response at selected wavelengths
     (dataset a)
         assuming multivariate normal distribution
         quadratic discriminant rule
          classes with different covariance structure
     testing the discriminant function: classification of
     spectra from dataset b



Katholieke Universiteit Leuven
                                 Laboratory measurements


  Spectral response of beet & weeds




Katholieke Universiteit Leuven
                                 Laboratory measurements

Spectral response of maize & weeds




Katholieke Universiteit Leuven
                                 Laboratory measurements

Spectral response of potato & weeds




Katholieke Universiteit Leuven
                                 Laboratory measurements


              Classification results




Katholieke Universiteit Leuven
                                            Field measurements


Field measurement of crop and weeds
                     Signal path
                                            Processing
                                            method

   Variation in
   light condition            Detector
                              sensitivity

                                   Measurement
                                   geometry
                                      Field measurements


   Equipment for field measurement

                                 spectrograph + 10-bit
                                 CCD, digital camera,
                                 computer,
                                 12 V battery and
                                 transformer

                                 on mobile platform




Katholieke Universiteit Leuven
                                                Field measurements


       Equipment - Spectrograph




          both spatial and spectral information in images

Katholieke Universiteit Leuven
                                                   Field measurements


                        Image data
     maize, sugarbeet,
     11 weeds
     2 different days,
     different light
     conditions
     755 x 484 pixels




                         spectral
                            axis
                                    spatial axis
Katholieke Universiteit Leuven
                                 Field measurements


         Spectral response of sensor




Katholieke Universiteit Leuven
                                      Field measurements


                   Data processing
     spectral resolution: 0.71 nm /pixel
     plant/soil discrimination with ratio: NIR
     (745 nm) / red (682 nm)
     data reduction by calculating average per
     2.1 nm, removing noisy ends
     resulting spectra: 484.8 - 814.6 nm range,
     2.1 nm step
     independent datasets of maize, sugarbeet
     and weeds

Katholieke Universiteit Leuven
                                     Field measurements


                 Spectral datasets




Katholieke Universiteit Leuven
                                 Field measurements

          Mean canopy reflections




Katholieke Universiteit Leuven
                              Field measurements

Canonical analysis of Sugarbeet - weeds
                           Field measurements

Canonical analysis of Maize - weeds
                                 Field measurements


    Discriminant analysis Sugarbeet




Katholieke Universiteit Leuven
                                 Field measurements


      Discriminant analysis Maize




Katholieke Universiteit Leuven
                      Field measurements

Graphic comparison datasets
                      Field measurements


Graphic comparison datasets
                                 Field measurements

    Graphic comparison datasets




Katholieke Universiteit Leuven
                        Field measurements

Discriminant analysis ratios
        Sugarbeet
                       Field measurements

Discriminant analysis ratios
          Maize
                                           Field measurements


                            Results
     only spectral info (485-815 nm)
     classification based on narrow bands in
     discriminant functions
         good results in similar light and crop
         conditions
         large decrease in performance for other
         light conditions
     using ratios of narrow bands
         improvement, but not sufficient

Katholieke Universiteit Leuven
                                                 Field measurements


                 Improving results
     influence of light conditions
         adaption of classification rule
             determining light condition and applying
             appropriate calibration/LUT
         spectral inputs that are less affected by
         environment
             measuring irradiance, calculating reflectance
         other classification methods



Katholieke Universiteit Leuven
                                            Crop-weed classification


    Neural network for classification
     Comparison of different NN techniques for
     classification
     Self-Organizing Map (SOM) neural network
     for classification
         used in a supervised way for classification
         neurons of the SOM are associated with local
         models
         achieves fast convergence and good
         generalisation.



Katholieke Universiteit Leuven
                                                                          Crop-weed classification


 Neural network for classification
           SOM                  MLP                                                 PNN

                                                                       x1 x2                xn-1 xn Input Layer
        Neural
                               class                                                              Distribution
        lattice                                                                                      Layer
          (A)
                           weights
                                          second hidden                                              Pattern Layer
                                              layer
Input                                ….
                                                  first hidden f1(x)                f (x)        Summation Layer
Space                                ….                                              2
                                                      layer
 (V)
                                                               Decision Layer
                                s2(k)
                            s1(k) s3(k)s4(k)                                         Output Layer
                                                                               O1                   On

   ADVANTAGES              ADVANTAGES                                   ADVANTAGES
   • Learns with reduced   • Good extrapolation                         • Fast Learning
   amounts of data         DISADVANTAGES                                • Retrainable
   • Fast Learning         • Slow Learning                              DISADVANTAGES
   • Visualisation         • Local minima                               • Needs all training data
   • Retrainable           • Needs a lot of data                        during operation
   DISADVANTAGES                                                        • Needs a lot of data
   • Discrete output
                                                         Crop-weed classification

  Comparison between methods




MLP: Multi-Layer Perceptron, PNN: Probabilistic N Network, SOM: Self-
Organizing Map, LVQ: Learning Vector Qantization, LLM: Local Linear Mapping
Moshou et al., 1998, AgEng98, Oslo
Moshou et al., 2001, Computers and Electronics in Agriculture 31 (1): 5-16
                                                             Crop-weed classification

      Comparison between methods




    MLP: Multi-Layer Perceptron, PNN: Probabilistic N Network, SOM: Self-
    Organizing Map, LVQ: Learning Vector Qantization, LLM: Local Linear Mapping



Katholieke Universiteit Leuven
                                                             Crop-weed classification

      Comparison between methods




    MLP: Multi-Layer Perceptron, PNN: Probabilistic N Network, SOM: Self-
    Organizing Map, LVQ: Learning Vector Qantization, LLM: Local Linear Mapping



Katholieke Universiteit Leuven
                                                  Crop-weed classification

           Conclusions on LLM SOM
                  technique
  • The strongest point is the local representation of the data
  accompanied by a local updating algorithm
  • Local updating algorithms assure much faster convergence than
  global updating algorithms (e.g. backpropagation for MLPs)
  • Because of the topologically preserving character of the SOM,
  the proposed classification method can deal with missing or noisy
  data, outperforming “optimal” classifiers (PNN)
  • The proposed method has been tested and gave superior
  results compared to a variety statistical and neural classifiers




Katholieke Universiteit Leuven
                                            Precision treatment

Precision spraying through controlled
           dose application
Unwanted variations in dose caused by horizontal and
vertical boom movements
                                                                   Precision treatment

    Active horizontal stabilisation of
              spray boom
      Validation with ISO 5008 track
          movement of spray boom tip with and without controller
                                                0.4

                                                0.3

                                                0.2



                                 Distance (m)
                                                0.1

                                                  0

                                                -0.1

                                                -0.2

                                                -0.3
                                                   0   5   10     15       20   25   30
                                                                Time (s)


Katholieke Universiteit Leuven
                                                                Precision treatment


  Vertical stabilisation of spray boom
  Slow-active system for slopes
           electric motor      reduction         fixing between plates
                                            g
                                                    frame connected to tractor

                                           rol
                                                                   q

                                                 boom
                                    cable


                            ultrasonic sensors

                              Resulting boom
                              movement


Katholieke Universiteit Leuven
                                              Precision treatment

   On-line selective weed treatment




        Indoor test of on-line weed detection and treatment
Katholieke Universiteit Leuven
                                              Precision treatment

         Indoor test of on-line weed
           detection and treatment
    Sensor: Spectral line camera
    Classification: Probabilistic neural network
    Program in Labview with c-code
        Image acquisition frequence: 10 images/sec, travel
        speed: 30cm/sec, segmentation with NDVI ( > 0.3)
        Off-line training of NN, On-line classification
        Decision to spray:
        > 20 weed pixels and > 35% of vegetation is weed
    Spray boom with PWM nozzles and controller,
    provided by Teejet Technologies

Katholieke Universiteit Leuven
                                 Precision treatment

    Indoor test of on-line weed
      detection and treatment
Color image and spectral image
                                    Precision treatment


              Indoor test - Results
     Comparison of nozzle activation with
     weed positions




Katholieke Universiteit Leuven
                                                      Precision treatment


              Indoor test - Results
Experimental set up                             - separate weed
                                       camera   classes (4) did not
                                                improve crop-weed
                              nozzle
                                                classification
                                                -Correct detection of
                                                nearly all weeds
                                                - Only 6 % redundant
                                                spraying of crop
                       weed
                                                - Up to 70 %
                                                reduction of herbicide
                                                use
Katholieke Universiteit Leuven

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:0
posted:8/5/2013
language:English
pages:46
vivien renata vivien renata
About