28-9-2014 Agrofood Robotics and Automation: From Farm to Table IEEE RAS TC on Agricultural Robotics and Automation Webinar #021 (AgRa) Rick van de Zedde Wageningen UR – 26 th of September 2014 Introduction Rick van de Zedde, business developer/ project leader for 10 years at Wageningen University & Research centre in The Netherlands Background: Artificial Intelligence, University of Groningen. Focus: computer vision/ robotics Contact: rick.vandezedde@wur.nl 1
28-9-2014 Wageningen University & Research Centre A university plus R&D organisations, mission statement; “ To explore nature and to improve the quality of life”. Wageningen UR: ● 6.500 employees ● 8.000 students ● 1.900 PhD’s ● 106 countries Wageningen UR - campus Wageningen UR - GreenVision ● GreenVision - the Wageningen UR centre of expertise on computer vision. ● Introduce new technology / scientific novelties into the agrifood industry together with industrial partners. ● 25 computer vision researchers within Wageningen UR. Coordinated by: Rick van de Zedde, Erik Pekkeriet, Gert Kootstra and Gerrit Polder ● One of the largest computer vision research groups in the (Dutch) agrifood industry. http://greenvision.wur.nl Contact: rick.vandezedde@wur.nl 2
28-9-2014 Outline Food inspection: from farm to table Farm/ breeding/ phenotyping Post-harvest quality inspection Packaging and robotics Retail Future perspective of R&D Contact: rick.vandezedde@wur.nl Outline Food inspection: from farm to table Farm/ breeding/ phenotyping Post-harvest quality inspection Packaging and robotics Retail Future perspective of R&D Contact: rick.vandezedde@wur.nl 3
28-9-2014 Mechanical intra-row weeders Erik Pekkeriet (PL), Pieter Klop, Jochen Hemming, ea. MARVIN – 3D based seedling sorting Rick van de Zedde (PL), Gerwoud Otten, Franck Golbach, ea. 4
28-9-2014 Specs, Scale and speed Current capacity: 19.500 seedlings/ hour is 185 ms/ seedling 10 GigE industrial machine vision camera’s, hardware triggered. 3D reconstruction technique: shape-from-silhouette/ volumetric intersection which requires a very accurate 3D camera calibration. Software runs on a fast Windows 7 desktop computer; National Instruments Labview/ core engine using CINs (C/C ++ ). Database: SQL server/ .Net web-interface Raw data collection issue - ±5 seedlings per second = 10 camera’s x 5 per second x 1.2MB per image = 60 MB / second 216 Gb / hour .... saving 3D models only = 0.5 Gb / hour 5
28-9-2014 Plant phenotyping and food production Challenge: produce food for 9 billion in 2050 Focus on improvement of crops (maize, rice, potatoes, tomatoes, etc). Novel genotyping technologies ‘deliver’ new varieties much quicker; ● Faster and less expensive DNA sequencers ● More efficient breeding cycles (GM and ‘classical’) Plants still need to be grown to determine yield, resistance against heat/drought stress, diseases , etc. (= phenotyping). So an increase of capacity/ objectivity is required: Opportunity for automated inspection, robotics, big data. Contact: rick.vandezedde@wur.nl Grant Agreement No. 284443. European Plant Phenotyping Network (EPPN) – 5.5 M € Goals: 1. Create a European integrated network/ community 2. Offer trans national access to EPPN facilities 3. Research – a. Novel sensors, b. Good practice phenotyping c. IT for high throughput Website: www.plant-phenotyping-network.eu Wageningen UR is WP leader of WP3 Novel sensors 6
28-9-2014 7
28-9-2014 8
28-9-2014 Vision-guided robotics Goal: propagating roses from cuttings 3D reconstruction and plant architecture. Rick van de Zedde (PL), Sanja Damjanovic, Gerwoud Otten, ea. 9
28-9-2014 Rose robot – automated planting Wageningen UR – Phenobot Gerrit Polder (PL), Fred van Eeuwijk, Marco Bink, Gerie van der Heijden, ea. 10
28-9-2014 Outline Food inspection: from farm to table Farm/ breeding/ phenotyping Post-harvest quality inspection Packaging and robotics Retail Future perspective of R&D Contact: rick.vandezedde@wur.nl Post-harvest quality inspection Automation in fruit/ vegetable production is widely used. ● Apple, oranges, mango’s, etc. using optical sorters. Individual products are analysed and graded. Common inspection method: several images of one product while continuously rotating on ‘wheels’. Alternative approach: 3D reconstruction and shape analysis developed for bell peppers (irregular shapes) Contact: rick.vandezedde@wur.nl 11
28-9-2014 Laser triangulation: shape measurements Contact: rick.vandezedde@wur.nl 3D shape analyses – multiple views Result based on one view Result of 6 sides Feature calculation like: Shape - identify middle of lobes Shape analysis: Block, point or ... shape Contact: rick.vandezedde@wur.nl 12
28-9-2014 3D shape analyses of bell peppers Quality criteria: ● Length/width ● Diameter ● Number of lobes ● Shape regularity ● Curvature NB: additional sensors required for colour/ defects. Contact: rick.vandezedde@wur.nl Lettuce handling with robots Kolen orienteren video hier toevoegen Contact: rick.vandezedde@wur.nl 13
28-9-2014 Outline Food inspection: from farm to table Farm/ breeding/ phenotyping Post-harvest quality inspection (2) Packaging and robotics Retail Future perspective of R&D Contact: rick.vandezedde@wur.nl Optical bulk sorting Contact: rick.vandezedde@wur.nl 14
28-9-2014 Video - impression Franck Golbach (PL), Gerwoud Otten, Roeland v. Batenburg, ea. Recording quality in high-speed mode Contact: rick.vandezedde@wur.nl 15
28-9-2014 Specs Detection and reject of products based on ● Colour defects (brown/ yellow/ green spots) ● Shape (length, width, curvature, complex shapes). Capacity: ● Conveyor belt speed up to 5 m/s ● 10.000 objects/ second, monitored with 4 camera’s. ● 20 - 30 tons per hour (with French fries) = 3 truck loads 4 linescan camera’s (2k pixels)/ Matrox Solios eCL framegrabbers/ PC-cluster with windows 7 plus Linux Patented ‘intelligent puffing’ product removal– shape based reject Contact: rick.vandezedde@wur.nl Hyperspectral / NIRS Measurement of quality aspects such as: ● Diseases/ product quality ● Moisture/ starch content ● Foreign materials in bulk streams Non-destructive and very fast (1 – 25 ms) Hardware: ● Spectrometers (260 – 2500 nm) ● Hyperspectral line-scan NIR camera for bulk sorting applications. Contact: rick.vandezedde@wur.nl 16
28-9-2014 Hyperspectral NIR camera Near InfraRed linescan camera (940 – 1790 nm) ● Xenics XEVA-343 xc104 - Specim N17E30 μm slit. ● Spectral resolution: 256 pixels - 3.3 nm/ pixel. ● Spatial resolution: 320 pixels. ● 100 frames (lines)/ second. NB: multispectral RGBi camera’s have colour channels from 400-700 nm and a near-infrared (NIR) channel at 750-900nm. Contact: rick.vandezedde@wur.nl Hyperspectral imaging Consider hyperspectral imaging when product and defects have: ● No density difference (no x-ray) ● No clear colour differences (no RGB) ● Quantitative measurements (ie. moisture content, fat content, inner decay) Warnings: ● Expensive hardware - InGaAs sensor instead of CCD/CMOS. ● Sensitive hardware - calibration/ humidity/ damage. ● Training set should cover all ‘real - life’ occurrences. NB: ● Seasonal differences will change the product ● Robust for several varieties of the product. ● Sensitive for the relevant range within product? Contact: rick.vandezedde@wur.nl 17
28-9-2014 Outline Food inspection: from farm to table Farm/ breeding/ phenotyping Post-harvest quality inspection Packaging and robotics Retail Future perspective of R&D Contact: rick.vandezedde@wur.nl Food processing industry Food processing factories have to be flexible: ● Large number of products and packaging variations ● Small batches ● Retailers place theirs orders late Contact: rick.vandezedde@wur.nl 18
28-9-2014 Food processing industry To meet increasing demands: ● Enormous amount of manual labour ● People are flexible Robots/machines are not flexible (yet) Contact: rick.vandezedde@wur.nl EU-project - PicknPack Large-scale EU funded research project Coordinator: Wageningen UR (Erik Pekkeriet) Budget: 14 M € Reducing manual labour in quality assessment and Consortium: packaging of food products. 14 universities, research institutes, companies incl. retail. www.picknpack.eu Contact: rick.vandezedde@wur.nl 19
28-9-2014 Flexible robotic systems for automated adaptive packaging of fresh and processed food products Sensing module Robot module Packaging module PicknPack – demonstrator PicknPack demonstrator Pick-and-place demo of vine tomatoes Dedicated gripper 20
Recommend
More recommend