International Dairy Nutrition Symposium, Wageningen, 16 October 2014 Strategies Towards a Quota-Free Dairy Production

Transcription

1 # Farms above 100 cows Agricultural Research Organization (ARO) Israel Animal Smart Sensing Agricultural Engineering Institution Trends in animal husbandry Large farms Ilan Halachmi The Volcani Centre, ARO, Bet Dagan. Israel Dairy Cattle Nutrition 16 October 2014 Wageningen Smart sensing ALLOW caring the individual animals These new trends require management skills and management concept. Trends in animal husbandry (zoom in) From Small family-run to Large farms with thousands of animals means: From Data to Results and Revenue Revenue Decision / Execution ACTION BENEFIT From Cow Names to Cow Numbers. Analysis KNOWLEDGE --> modeling From Owner Operated to Employees & Organizations Big influence of product prices and feed costs. Integration INFORMATION Animal Health and Animal-welfare branding Large companies become bigger Collection DATA (sensors) COST Tel: Ilan Halachmi Smart Sensing Concept Spreading sensors at key locations of the Farm Analysis and reports (Management by exception) Modeling / decision support systems Agricultural Research Organization ARO Israel Amos Bercovich Y. Edan, V. Alchanatis, E. Maltz, I. Halachmi Automatic body condition scoring

2 Radial distance from object center to contour Automatic Body Condition Scoring (BCS) Body condition scoring (BCS) estimates energy reserves (fatness) of cows Important management tool health issues Optimal time for 1 st insemination feed management The Problem Manual Hard work Labor & Time consuming Subjective Technician Previously seen cows Thin Fat 7 halachmi@volcani.agri.gov.il Automatic body condition scoring Manual labeling of 23 anatomical points Bewley et al., 2008 Azzaro et al., 2011 Polynomial fitting to cow contour Thermal camera Halachmi et al., Automatic body condition scoring Algorithms-segmentation אילן הלחמי Tel: Ilan Halachmi Automatic body condition scoring Cow contour: Interpolation 1000 points Scaling to 0-1 range 5 anatomical points Horizontal and vertical distances 5 angles 1 dimension curve 11 Algorithms Fast Fourier transform Thin cow (BCS < 2.5) Fat cow (BCS > 3.5) Feature Extraction Contour index

3 Model Output Model output BCS sensor 1.5 Results (Training database :87 cows) R 2 = Model input-manuall BCS Gold Reference - Manual BCS BCS Sensor Regular camera (RGB) Model Repeatability Image Processing and Signal Processing Applied Results (1-5 Scale) Range 0-1/4 0-1/2 0-3/4 0-1 Training set 53% 82 % 98% 100% Testing set 43% 72 % 94% 100% 2013 JDS Outline Automatic body condition scoring Automatic cow lameness detection Approaching calving sensors Automatic early detection of calving diseases Mega Dairy in the far east Automatic lameness detection case study 2 Tom van Hertem s Ph.D. Marie Curie EU-FP7 BioBusiness project Netherlands Belgium Sweden 18 Two combined ways COMPUTER VISION Automatic LAMENESS detection BEHAVIOUR SENSING Israel 2.5 M 11 PhD Students Decision making system

4 Model Step 1: selecting key indicators Could use: Lying behavior, Rumination, Neck Activity, Body Weight, Milk components. etc Automatic cow lameness detection computer vision 19 Picture taken in Yefat, ARO cows אילן הלחמי Tel: Ilan Halachmi 3 rd Setup: 3D-camera 21 Camera Camera 22 Behavior Sensing SCR HR-Tag Ruminating time [min/2h] Activity [activity index/2h] Antena SCR Milk meter ARO, The Volcani Centre - 23 May 2013 Model Development Logistic regression: 7 input variables: Milk: Daily yield, week slope, ratio and difference Neck activity: daily sum, day/night ratio Rumination: only night Validation Reference 0.86 Lame Not-Lame Lame Not-Lame ISAE, Bet Dagan - 23 May 2013 Model Development Classification Tree Dichotomous question (yes/no) at each node Cope with missing data Intuitive results Milk <30 Camera Rumination<150 Lame Not- Not-Lame Lame Activity>500 Lame

5 Model Score 25 Algorithm Verification Sensitivity Specificity Accuracy Calibration Verification N=744 cows Algorithm Verification Reference Score 90.9% % Sensitivity Specificity Accuracy = ability to detect lame animals = ability to detect not-lame animals = ability to detect lame and not-lame animals (correct classification rate) ISAE, Bet Dagan - 23 May Tom van Hertem ISAE, Bet Dagan - 23 May 2013 Lameness Sensor Rumination, Neck Activity, Milk, 3D-Camera The Decision Model Logistic Regression and Classification Tree Applied Results TBD at the end of the project. ARO library Outline Automatic cow lameness detection Approaching calving sensors Automatic early detection of calving diseases Mega Dairy in the far east Automatic early detection of calving diseases Machteld Steensels s Ph.D. Which sensors? 30

6 Lying time (min/day) Data collection 4 farms 1 x 1100 cows/farm 3 x 300 cows/farm Which model? Logistic regression model Probability of being sick f(z) = (1 + e-z)-1 z = β0 + β1 x1 + β2 x2 + + βk xk Model outcome: 0 Healthy : 1 Ketotic Variables: Rumination Time, Activity and Milk Yield Model Calibration Coefficients in the validation phase: On 1 farm to others Or On a % of farm data (within a farm) 2/3; Leave one out, etc equal number of healthy and sick cows Lying time Lying is an important issue to animal welfare What about health? 33 Dr. Ilan Halachmi halachmi@volcani.agri.gov.il Preliminary results: Lying behavior Health cows 550 Runimation Healthy vs. ketotic cows Neck Activity Sick cows Days after calving (days) 143 sick cows 140 healthy cows, Summer 35 Milk 36

7 VALIDATION Rumination (minutes/day) Day 2 Day 3 Day 4 Day 5 Day 6 Healthy 348± ± ± ± ±4.8 Light 308± ± ± ± ±7.5 Moderate & Severe Mean ± SE Relative to claving day 260± ± ± ± ±22.4 Model: Validation Focus > Hypothesis > Objectives > M&M > Preliminary Results > Model> Future Research Farm Correct 1 83 % 2 70 % 3 91 % 4 67 % 5 77 % Machteld Steensels ISAE, Bet Dagan - 23 May Model accuracy CALIBRATION Farm 1 Farm 2 Farm 3 Farm 4 Farm N tot Acc. Acc. Acc. Acc. Farm 1 Ket Heal 309 Farm 2 Ket Hea 46 Farm 3 Ket Hea 61 Ketosis: Many Sensors Lying behavior, Rumination, Neck Activity, BW, Milk components Logistic Regression Decision Model Applied Results Ref: Machteld s PhD, in the ARO library Farm 4 Ket Hea Automatic early detection of calving diseases Application in a robot Outline Automatic cow lameness detection Approaching calving sensors Automatic early detection of calving diseases Mega Dairy in the far east

8 Maximal activity (steps/h) Lying time (h) Lying sensor (Leg angle) Sensing approach calving - the Algorithm 1. Lying time down 2. Number of steps up Ephraim Maltz ( Afimilk Afimilk Behavior 1 Behavior 2 Direction of the daily change Sensing an approached calving The Ratio: Steps/lying= restless index activity lying time Practical Application Behavior 2 Diurnal Practical Application Days Behavior Days before calving Approached calving - two Sensors Steps, Lying sensor (leg angle) Decision Model Multi Variables Regression ~ restless Applied Results 76% Maltz and Antler. CIGAR2008 Brazil Outline Automatic cow lameness detection Approaching calving sensors Automatic early detection of calving diseases Mega Dairy in the far east

9 A mega dairy in India in brief Design & management of a large-scale dairy farm require OR tools. In this study a combined model: queuing-network, robust 6σ design, simulation and optimization was developed Systems engineering Agricultural Research Organization (ARO) Israel Design all components as one single system Design criteria were: 10,000 cows in milking, intensive farming while maximized animal welfare, year-round indoors, no grazing, open-large cowsheds, Simulation dry manure bedding, & no cubicle housing, maximizing cow resting time and worker convenience. All design criteria were Optimization met. We modeled eight farming aspects: * cow traffic, * milking parlors, * vet treatment, * manure handling, * cow cooling, * feed-center operation, * workers' transportation * a problematic junction, and their interrelations. Tel: Ilan Halachmi אילן הלחמי Outline Automatic cow lameness detection Approaching calving sensors Automatic early detection of calving diseases Mega Dairy in the far east The right key indicators Take Home message Our farm has many sensors Automatic data collection Cow individual History is of importance Data processing & machine learning Management Concept Meeting outline Automatic early detection of calving diseases automatic cow lameness detection approaching calving sensors Mega Dairy halachmi@volcani.agri.gov.il