Changying “Charlie” Li, Ph.D. Associate Professor University of Georgia AgRa Webinar October 24, 2013
E-nose Fluorescence imaging of plants and cotton trash Multi- sensor platform Berry Impact Recording Device // Monte Carlo algorith main() 0 1 0 0 1 0 1 0 0 1 0 0 1 { char str[STRLEN]; 1 1 0 1 1 0 1 1 0 1 1 0 0 sprintf(str,”Parallel 0 1 0 0 1 0 1 1 0 1 0 0 1 printf(“%s\n”, str); 1 1 1 0 0 0 1 1 1 1 0 1 1 Parallel_Run(); 0 1 0 0 1 0 1 0 1 1 0 0 0 0 0 0 1 0 0 1 1 0 0 0 0 1 Results_Processing(); 0 1 0 0 1 0 1 1 0 1 0 0 1 ........... 1 1 0 0 1 0 1 1 1 1 0 0 0 0 1 0 0 1 0 1 0 0 1 0 0 1 2
Intelligence: learning, planning, navigation Sensing and perceptions Mobility and manipulation 3
Hyperspectral imaging for onion quality inspection Electronic nose for rotten onion detection in storage Berry Impact Recording Device for blueberry mechanical harvester improvement 4
Onion grading robot Developing a nose for robots A BerryBot to diagnose harvesters 5
Onion grading robot Haihua Wang (former Ph.D. student) Wang, H., C. Li, and M. Wang. 2013. Quantitative determination of onion internal quality using hyperspectral imaging with reflectance, interactance, and transmittance modes. Transactions of ASABE. 56(4): 1623-1635. 6
I. SCRI Onion Postharvest Projects Advancing Onion Postharvest Handling Efficiency and Sustainability by Automated Sorting, Disease Control, and Waste Stream Management USDA competitive grant: Specialty Crops Research Initiative ($774,581) Multi-state, comprehensive 4-year research/ extension project to take onion postharvest handling to next level 7
Onion is the largest vegetable in GA and third largest in the U.S. ($1 billion) 13% of the total onion production in the U.S. goes to dehydration and processed market Internal quality (e.g., dry matter) is important Nondestructive sensing methods are not available for onion industry. http://www.baldorfood.com 8
Easily get fatigued Fail to detect internal defects and latent fungal diseases Labor intensive and high cost (50%) Unable to evaluate internal quality properties 9
Refractometer (SSC) Magness ‐ Taylor testing platform (Firmness) Oven (DM) 10
Pixel spectra at (x,y) Reflectance Wavelength ( λ ) Spectral imaging Spectroscopy - external defects sugar content prediction detection for apples, cantaloupes, - diffuse reflectance prune, papaya, tomatoes - none for onion Birth et al. 1985: onion 11
12
13
2 2 2 X Y Z 1 2 2 b a R R 0 cos N P y 0 cos N P 0 2 2 2 x y z 2 2 2 a ( j x ) a ( j x ) ( j x ) 0 0 0 N P P 1 2 [ , , ] 2 2 b d ( i z ) b ( i z ) d 14 0 0
15
Reflection Interaction Transmission 16
17
18
This study proved efficacy of hyperspectral • imaging for onion internal quality prediction. Interactance mode can be used to reliably predict • SSC and DM of onions. Next step: implement interactance in packing • lines 19
Let the robot have a nose Tharun Konduru (former M.S. student) 20
Annual production and storage losses in onion as a result of diseases can reach 50% or more; Botrytis neck rot (caused by the fungus Botrytis allii ) and sour skin (caused by the bacterium Burkholderia cepacia ) are most serious threats. Botrytis neck rot Sour skin 21
22
Develop a customized and low cost gas sensor array (E-nose) ◦ Mechanical ◦ Electronic ◦ Software Test the sensor for sour skin disease detection in onions 23
7 MOS sensors + Temp + RH sensors Temp/RH Pump sensors Teflon chamber Exhaust Gas sensors Gas inlet Clean Valve air inlet 24
25
26
27
28
• Sample preparation: jumbo yellow onions were bought in local store; surface sterilized • Inoculation and incubation: Burkholderia cepacia , strain Bc 98-4; 1mL of bacterial inoculum was injected on two opposite sides of the neck region of the onion (~30mm deep) X 8 X 8 29
Batch 1 Batch 2 Control Diseased Control Diseased 3 rd dai 16 16 24 24 4 th dai 16 16 24 24 5 th dai 24 24 23 23 6 th dai 24 24 24 24 7 th dai 24 24 24 24 Total 104 104 119 119 Total = 446 30
Diseased Healthy 31
32
LDA SVM All S2,S3,S4 S2,S3,S4 All S2,S3,S4, S2,S3,S4 S5,S6,S7 S6,S7 S5,S6,S7 S6,S7 B1->B2 81.58 56.3 43.15 85.26 81.05 75.78 Average 88.24 86.26 87.63 91.8 92.34 92.36 Leave-1-out 89.89 88.5 88.52 92.35 91.53 92.62 SVM is better than LDA Two cross validation methods were better than B1->B2 Sensor reduction to 5 could be achievable. 33
A low cost gas sensor array was successfully developed with an automated gas delivery system and data acquisition features Validation tests showed that the device can differentiate sour skin infected onions from healthy onions starting from four days after inoculation. The sensor has the potential to be used for onion disease detection in storage. 34
Rot onion tracing in a large storage room Concentration 0 12.5 25 37.5 50 62 75 87.5 100 mg/kg 35
BerryBot to diagnose machine harvesters Development of a Smart Blueberry Funded by SCRI blueberry mechanical harvest project Pengcheng Yu (Former M.S. student) 36
Blueberry Mechanical Harvester Rotary harvester 37
Berry Impact Record Device (BIRD) Overall goal: to develop an “instrumented sphere” sensor to measure impacts, identify sources of bruising and optimize mechanical harvesters (1) BIRD Sensor node (2) BIRD Interface box (3) PC ‐ BIRD Software (4) DC Power supply 38 for the interface box
BIRD Sensor 39
40 BIRD at Work
Blueberry Mechanical Harvest Field Test
Real Time Impacts (Rotary) 600 500 Phase 1 Phase 2 400 500 Phase 3 300 Impact (g) Phase 4 200 400 100 Impact (g) 0 300 0.696 0.698 0.700 0.702 0.704 0.706 0.708 Time (s) 200 100 0 0.7 0 2.2 4 6.9 7.3 6 Time (s) 42
Sensor design met design criteria: Size (25.4 mm) Frequency (3 kHz) Memory (1 MB) Battery (2.5 h) Sensing range (500g) Accuracy (0.53%) Cost ($350) Field test: Quantitatively measures impacts during mechanical harvesting (rotary) Identified critical control points 43
Collaborators Students, postdocs, visiting scholar, technician. 46
45
Thank you! cyli@uga.edu http://sensinglab.engr.uga.edu 46
Recommend
More recommend
