High dependency on fossil fuels Whats next? High dependency on - - PowerPoint PPT Presentation

▶

Aug 31, 2022 354 likes •576 views

First agricultural revolution ~12000BC Second agricultural revolution 18th&19th Century Current situation High dependency on fossil fuels Whats next? High dependency on chemicals Sustainable Agriculture Advanced

SLIDE 1

SLIDE 2

First agricultural revolution – ~12000BC Second agricultural revolution – 18th&19th Century Current situation

High dependency on fossil fuels
High dependency on chemicals
Advanced genetics for increased crop yield
Fast growing human population

What’s next? Sustainable Agriculture

SLIDE 3

Weed control is a fundamental operation for any crop to maintain high yields 90% is performed chemically Large machinery

Oil consumption
CO2 emission

Weeding alternatives

Mechanical weeding (Hoeing machine) – Not efficient
Manual weeding – 10 times more expensive

SLIDE 4

Untargeted application…

>90% wasted → high costs
Toxic residues on soil, water, crop → health & environment impact
Damage on crops → 5-10 % yield losses in average

…of selective herbicides

New molecules expensive to develop ($350M1)
Increased legal pressure to ban molecules & reduce quantity
Development of herbicide-resistant weeds (+10%/year2)

1 Phillips McDougall, 2015 2 “Global Increase in Unique Resitant Cases”, Dr Ian Heap, Weedscience.org, 2016

SLIDE 5

Autonomous weeding / navigation

Minimal human intervention

Solar powered

Sustainable energy & Power autonomy

AI for weed detection Robotic arms for precision spraying

Same efficiency up to 40% cheaper
No chemicals on crops
Up to 20 times less herbicide
Allows cheaper & ecological molecules
Reduces herbicide-resistant weeds problem

SLIDE 6

SLIDE 7

Solar panels, 350 W max Delta robotic arms Electric motors, 0.5 m/s speed GPS & IMU for navigation RGB Camera, 5 MP @10 fps Jetson TK1 (TX2) for CV & ML Weed detection @ >1 fps

SLIDE 8

Row Finder Pipeline Image Downscaling Plants Segmentation Row Finder Weed Detect Pipeline Plants Segmentation Feature Extraction Classification Post-processing Map weed positions to robot coord. Send weed positions & sizes to arms Camera Settings Adjustment & Image Acquisition

SLIDE 9

Row Finder Pipeline Image Downscaling Plants Segmentation Row Finder Weed Detect Pipeline Plants Segmentation Feature Extraction Classification Post-processing Map weed positions to robot coord. Send weed positions & sizes to arms Camera Settings Adjustment & Image Acquisition

SLIDE 10

JETSON TK1 JETSON TX2 CPU GPU CPU GPU 424.6 ± 18.1 55.6 ±1.86 130.9 ± 27.9 36.7 ± 10.8 107.2 ± 2.6 0.41 ± 1.21 79.9 ± 15.7 0.21 ± 0.71 208.9 ± 6.1 41.3 ± 8.9 196.7 ± 6.4 14.7 ± 4.92 881.3 ± 20 98.45 ± 9.1 517.3 ± 42.4 51.8 ± 14.1 Plant Pre-Segmentation Color Space Conversion Color Normalization Gaussian Smoothing Overall

(Mean Computation Times over 90 images in ms)

SLIDE 11

PROTOTYPE 1 PROTOTYPE 2

HW: JETSON TK1 HW: JETSON TX2 Hand-crafted Features DNN AdaBoost Classifier DNN

SLIDE 12

SLIDE 13

PROTOTYPE 1 PROTOTYPE 2 HW: JETSON TK1 HW: JETSON TX2 Hand-crafted Features DNN AdaBoost Classifier DNN OA = 92% F1=0.87

Prec. = 97.2% Rec. = 78.3%

OA = 92.2% F1=0.867

Prec. = 96.3% Rec. = 78.8%

OA = 97.8% F1=0.77

Prec. = 94.4% Rec. = 65.6%

OA = 97.8% F1=0.78

Prec. = 87.3% Rec. = 70.5%

Set 1 – Seen field (w.r. 33.9%) Set 2 – Seen field (w.r. 5.6%)

OA = Overall Binary Accuracy, F1 = Mean F-Score

Prec. = Weed detection precision,
Rec. = Weed detection recall

w.r. = Weed / (Weed + Crop) pixels ratio

SLIDE 14

Effects of Shadow on Color & Visibility

SLIDE 15

Illumination Changes due to daytime & weather 10am 6pm

SLIDE 16

Soil & Crop Variation Field 1 Field 2

SLIDE 17

PROTOTYPE 1 PROTOTYPE 2 HW: JETSON TK1 HW: JETSON TX2 Hand-crafted Features DNN AdaBoost Classifier DNN OA = 79.4% F1=0.531

Prec. = 56.7% Rec. = 49.9%

OA = 92.5% F1=0.863

Prec. = 86.5% Rec. = 86.1%

Mixed Set – Unseen Fields (w.r. : 23.4%)

OA = Overall Binary Accuracy, F1 = Mean F-Score

Prec. = Weed detection precision,
Rec. = Weed detection recall

w.r. = Weed / (Weed + Crop) pixels ratio

SLIDE 18

JETSON TK1 –

Feat. Extraction + AdaBoost

JETSON TX2 –

Feat. Extraction + Adaboost

Jetson TX2 – Deep Learning (CNN) 744 ± 503.4 ms 634.5±393.5 ms 987.1 ± 523.7 ms After Optimization 780 ± 447.6 ms Full Pipeline 977.4 ± 511.9 ms Full Pipeline 833.8 ± 397.6 ms Full Pipeline 879 ± 457.5 ms (Mean Computation Time over 20 images)

SLIDE 19

Real-life embedded GPU application that can improve food production First completely autonomous weeding robot NVIDIA Jetson TX2 opens new frontiers for embedded platforms Deep Learning is powerful, but it is no magic wand Challenging to obtain images covering all situations

SLIDE 20

First agricultural revolution – ~12000BC Second agricultural revolution – 18th&19th Century Current situation

What’s next? Sustainable Agriculture

Weed control is a fundamental operation for any crop to maintain high yields 90% is performed chemically Large machinery

Weeding alternatives

Untargeted application…

…of selective herbicides

Autonomous weeding / navigation

Solar powered

AI for weed detection Robotic arms for precision spraying

Solar panels, 350 W max Delta robotic arms Electric motors, 0.5 m/s speed GPS & IMU for navigation RGB Camera, 5 MP @10 fps Jetson TK1 (TX2) for CV & ML Weed detection @ >1 fps

Row Finder Pipeline Image Downscaling Plants Segmentation Row Finder Weed Detect Pipeline Plants Segmentation Feature Extraction Classification Post-processing Map weed positions to robot coord. Send weed positions & sizes to arms Camera Settings Adjustment & Image Acquisition

Row Finder Pipeline Image Downscaling Plants Segmentation Row Finder Weed Detect Pipeline Plants Segmentation Feature Extraction Classification Post-processing Map weed positions to robot coord. Send weed positions & sizes to arms Camera Settings Adjustment & Image Acquisition

(Mean Computation Times over 90 images in ms)

PROTOTYPE 1 PROTOTYPE 2

HW: JETSON TK1 HW: JETSON TX2 Hand-crafted Features DNN AdaBoost Classifier DNN

PROTOTYPE 1 PROTOTYPE 2 HW: JETSON TK1 HW: JETSON TX2 Hand-crafted Features DNN AdaBoost Classifier DNN OA = 92% F1=0.87

OA = 92.2% F1=0.867

OA = 97.8% F1=0.77

OA = 97.8% F1=0.78

Set 1 – Seen field (w.r. 33.9%) Set 2 – Seen field (w.r. 5.6%)

OA = Overall Binary Accuracy, F1 = Mean F-Score

w.r. = Weed / (Weed + Crop) pixels ratio

Effects of Shadow on Color & Visibility

Illumination Changes due to daytime & weather 10am 6pm

Soil & Crop Variation Field 1 Field 2

PROTOTYPE 1 PROTOTYPE 2 HW: JETSON TK1 HW: JETSON TX2 Hand-crafted Features DNN AdaBoost Classifier DNN OA = 79.4% F1=0.531

OA = 92.5% F1=0.863

Mixed Set – Unseen Fields (w.r. : 23.4%)

OA = Overall Binary Accuracy, F1 = Mean F-Score

w.r. = Weed / (Weed + Crop) pixels ratio

JETSON TK1 –

JETSON TX2 –

Jetson TX2 – Deep Learning (CNN) 744 ± 503.4 ms 634.5±393.5 ms 987.1 ± 523.7 ms After Optimization 780 ± 447.6 ms Full Pipeline 977.4 ± 511.9 ms Full Pipeline 833.8 ± 397.6 ms Full Pipeline 879 ± 457.5 ms (Mean Computation Time over 20 images)

Real-life embedded GPU application that can improve food production First completely autonomous weeding robot NVIDIA Jetson TX2 opens new frontiers for embedded platforms Deep Learning is powerful, but it is no magic wand Challenging to obtain images covering all situations

Contact: Anıl Yüce anil.yuce@ecorobotix.com