Seediscussions,stats,andauthorprofilesforthispublicationat:http://www.researchgate.net/publication/250719002

Extractingpoultrybehaviourfromtime-seriesweighscalerecords

ARTICLEinCOMPUTERSANDELECTRONICSINAGRICULTURE·JUNE2008

ImpactFactor:1.49·DOI:10.1016/j.compag.2007.08.015

CITATIONS

7

2AUTHORS,INCLUDING:

HongweiXin

IowaStateUniversity

288PUBLICATIONS1,690CITATIONS

SEEPROFILE

Availablefrom:RichardStephenGates

Retrievedon:27August2015

C

Ew

1

2

R3

a3

b4

5

a6

7

K8

P9

B10

A11

W12

F13

1

T14

m15

s16

r17

m18

d19

b20

i21

a22

t23

o24

c25

l26

a27

c28

o29

r30

Q1

1 02 d

PR

OO

F

ARTICLE IN PRESSOMPAG 2049 1–7

c o m p u t e r s a n d e l e c t r o n i c s i n a g r i c u l t u r e x x x ( 2 0 0 7 ) xxx–xxx

avai lab le at www.sc iencedi rec t .com

journa l homepage: www.e lsev ier .com/ locate /compag

xtracting poultry behaviour from time-serieseigh scale records

ichard S. Gatesa,∗, Hongwei Xinb

Biosystems and Agricultural Engineering Department, 128 C.E. Barnhart Building, University of Kentucky, Lexington, KY 40546, USAAgricultural and Biosystems Engineering Department, 3204 NSRIC, Iowa State University, Ames, IA 50011-3310, USA

r t i c l e i n f o

eywords:

oultry

ehaviour

nimal well-being

elfare

a b s t r a c t

Algorithms for determining individual bird feeding statistics and stereotyped pecking

behaviour from time-series recordings of feed weight were developed and compared to video

observations. Data taken from two separate experiments involving broiler and laying hen

chickens were used to evaluate the algorithms. The effects of algorithm tuning parameters

including thresholds for changes in weight and sequential number of stabilized readings,

D eeding arithmetic moving average for meal tare values, and the sampling frequency of feed weight

recordings were evaluated. Results suggest that a minimum sampling frequency of 0.5–1 Hz

is recommended for discerning behavioural changes that include timing of feeding events

and their duration. However, lower sampling frequencies are acceptable for determining

hourly (or greater) feed consumption.

31

32

33

34

35

36

37

38

39

40

41

42

43

44

45

mine variation in feed and water use amongst individual birds 46

OR

RE

CTE

. Introduction

he question of how management or environmental stimuliay influence poultry behaviour and/or well-being is of con-

iderable importance for fundamental studies of behaviouralesponse to stimuli, and as a means of assessing appropriate

anagement and environmental designs for commercial pro-uction. What responses should be measured and whetherird response is correlated to well-being are active areas of

nvestigation. If multiple choices or stimuli are available, it isconsiderable challenge to assign behavioural outcomes to

hese treatments. Discrimination between competing choicesr stimuli requires careful experimental design to assess birds’hoice selection. Wathes et al. (2001) list a set of criteria postu-ated by Abeyesinghe (2000), which can be used to “normalize”ssessments of animal response. These response assessment

UN

C

Please cite this article in press as: Gates, R.S., Xin, H., Extracting poultryAgric. (2007), doi:10.1016/j.compag.2007.08.015

riteria include a need for sensitivity to all stimuli, responsivever different time periods and levels of stimulus, and suitableepeatability for scientific assessment.

∗ Corresponding author. Tel.: +1 859 257 3000; fax: +1 859 257 5671.E-mail address: rich.gates@uky.edu (R.S. Gates).

168-1699/$ – see front matter © 2007 Elsevier B.V. All rights reserved.oi:10.1016/j.compag.2007.08.015

© 2007 Elsevier B.V. All rights reserved.

One means of assessing bird response to stimuli involvescareful analysis of characteristics of individuals or groupsover time. Monitoring individual behaviour during researchtrials is typically performed with some type of video imag-ing system. For poultry, behavioural activities are categorizedinto events such as eating, drinking, preening, resting,and stereotyped activities directed at different targets. Thisassessment methodology is time-consuming, hence costly,tedious and prone to errors, even with modern commer-cially available research systems that compile the statisticssemi-autonomously. There is an increasing need for meansto further automate collection of event-based behaviouralresponses (Gates et al., 1995; Gates and Xin, 2001; Persyn etal., 2003, 2004; Xin et al., 1993).

With behavioural monitoring, it is not feasible to deter-

COMPAG 2049 1–7behaviour from time-series weigh scale records, Comput. Electron.

within a treatment, nor do they dynamically monitor feed 47

and water intake for the same bird as environment is modi- 48

fied. Recent measurements with the Individual Bird Unit (IBU)

ED

PR

OO

F

IN PRESSCOMPAG 2049 1–7

a g r i c u l t u r e x x x ( 2 0 0 7 ) xxx–xxx

49

50

51

52

53

54

55

56

57

58

59

60

61

62

63

64

65

66

67

68

69

70

71

72

73

74

75

76

77

78

79

80

81

82

83

84

85

86

87

88

89

90

91

92

93

94

95

96

97

98

99

100

101

102

103

Fig. 1 – The IBU feed scale system installed in one chamberwith six cages (top). View of a single cage, with feed

104

105

106

107

108

109

110

111

112

113

114

115

NC

OR

RE

CT

ARTICLE2 c o m p u t e r s a n d e l e c t r o n i c s i n

system (Puma et al., 2001; Cook and Xin, 2004) indicate thatindividual birds adjust their eating and drinking behaviourquite differently for the same thermal treatment, and thatthis effect is masked when comparing group means. Collec-tion of data for variability between individuals, if practical,may provide an efficient basis for assessing bird responseusing population percentages, minimization (or elimination)of extreme responses, or genetic improvement by individualselection of previously unavailable selection criteria (Naas etal., 2000).

One set of behavioural assessment criteria is feeding activ-ity. Measures include number of meals, meal size, mealduration, ingestion rate, meal intervals, and proportion oftime spent eating. In addition, birds spend varying amountsof time pecking without eating, defined as stereotyped peck-ing behaviour. Such information may be useful to understandhow to better design housing systems to satisfy bird’s inher-ent needs for food, and to study the space requirementsand the impact of competition in commercial settings (Cookand Xin, 2004). Behaviour of individual birds at the feeder, ifquantified, could form a comparative basis for assessing alter-native management and housing strategies (Persyn et al., 2003,2004).

The objective of this research was to devise, test and val-idate two algorithms to determine individual bird activitiesincluding time at station, activity at station, meal size andduration, for use with time-series recordings of feed levelsfrom the existing Individual Bird Use system. In addition, onealgorithm was tested in its ability to extract time allotmentactivities as compared to video recordings. In the followingsections, we describe the two algorithms, discuss the effectof key parameters to tune the algorithms, and as appropriatecompare them to baseline data.

2. Materials and methods

2.1. Equipment

2.1.1. IBU systemThe Individual Bird Unit (IBU) system consisted of 24 feed-ing/drinking stations divided into four groups of six stations.Each group was located in one of two environmentally con-trolled chambers. Each feeding station (Fig. 1) consisted of aprecision electronic weighing scale (model CT1200, Ohaus Cor-poration, Florham Park, NJ) with a 1210 g capacity and a 0.1 gresolution and a plastic feeder measuring 13 cm (L) × 13 cm(W) × 15 cm (H) (5 in. (L) × 5 in. (W) × 6 in. (L)). The plastic feederhad a u-shaped access side opening and its bottom was fas-tened to the electronic scale with Velcrotm strips. Each scalehad an RS232 serial interface connected to a custom-builtmicrocontroller with RS232 and RS485 communication ports,digital input/output and analogue/digital converter (KG Sys-tems Inc., East Hanover, NJ). The 24 microcontrollers werenetworked to a master unit via the RS485 ports; the mastermicrocontroller assigned polling commands, collected infor-

U

Please cite this article in press as: Gates, R.S., Xin, H., Extracting poultryAgric. (2007), doi:10.1016/j.compag.2007.08.015

mation from each unit, and forwarded the data to a PC viaRS232. The weigh scales were located on a wooden stand infront of the individual birdcages. The cages measured 25 cm(W) × 46 cm (D) × 46 cm (H) (10 in. (W) × 18 in. (D) × 18 in. (H)).

116

weighing system.

Complete details of the IBU system can be found in Puma etal. (2001).

Readings of the feeder weighing scale were scanned onperiodic command and captured with a Visual Basic macroexecuting in MS Excel. Maximum sampling frequency forthe IBU system is one sample each 4 s (i.e. Ts = 4 s) withall 24 units operational. For purposes of the work reportedhere, sampling times Ts of 4 and 30 s were selectivelyused.

Behavioural data of two groups of four birds were acquiredwith a video recording system that consisted of two CCDcameras (Panasonic, AG-6730), a time-lapsed VCR (Panasonic,PV-V4520) and a TV monitor. The cameras were mounted so

COMPAG 2049 1–7behaviour from time-series weigh scale records, Comput. Electron.

that full images of four neighbouring hens could be recorded. 117

The recordings were compiled on an hourly basis and used 118

for comparison purposes. The number, duration and type of 119

INCOMPAG 2049 1–7

a g r

e120

r121

t122

a123

f124

2125

F126

o127

e128

T129

p130

(131

f132

o133

i134

C135

t136

s137

p138

w139

m140

t141

2142

B143

T144

145

146

147

148

149

150

151

152

153

154

155

156

157

158

159

160

161

162

163

164

165

166

167

168

ARTICLEc o m p u t e r s a n d e l e c t r o n i c s i n

vents (time at feeder, time at drinker, remainder of non-esting time) were tabulated from visual analysis of theime-lapsed recordings. Four hens during 1 day of heat stressnd two hens during 1 day of the recovery period were utilizedor this activity.

.1.2. High-frequency sampling systemour measurement stations were used in a separate study tobtain high-frequency sampling data. Each station had onelectronic balance (2000 ± 0.1 g) (model HF-2000, AND Inc.,okyo, Japan). The balance provided a 0–1 VDC analogue out-ut for the weight range. A rectangular aluminium feeder

20 cm (W) × 10 cm (D) × 5 cm (D)) was attached to the plat-orm of the balance using Velcro. The analogue output signalf each balance was connected to a differential analogue

nput channel of an electronic data logger (model CR23X,ampbell Scientific Inc., Logan, UT, USA). The CR23X con-

ained a 4 MB extended storage memory, and it measured andtored the output signals at Ts = 0.1 s intervals (10 Hz sam-ling frequency). At this sampling rate, about 30 MB of raweight and time data were collected per day for the foureasurement stations. These data were downloaded hourly

o a PC.

UN

CO

RR

EC

TED

Please cite this article in press as: Gates, R.S., Xin, H., Extracting poultryAgric. (2007), doi:10.1016/j.compag.2007.08.015

.2. Experimental birds

oth laying hens and broilers were utilized in these studies.he data from the IBU system was taken from a study on

Fig. 2 – Flow chart of main s

RO

OF

PRESSi c u l t u r e x x x ( 2 0 0 7 ) xxx–xxx 3

effects of drinking water temperature during diurnal, warm-to-hot environments (Xin et al., 2002). The hens were W-36layers, approximately 32 weeks old at the start of the test.Lights were turned on at 5:00 and off at 21:00 each day. Feedwas replenished daily between 8:10 and 8:50. A single dayof data from the 3rd week of a 4-week heat stress event(two hens), and 2 days from the 2nd week of a 2-week ther-moneutral recovery period (four hens) were available. Thesehen × day combinations were selected for analysis becausethey had few missing values and simultaneous video record-ings of these birds.

The high-frequency data were taken from a broiler feedingstudy (Japanese, Chunky breed) to assess feeding behaviourand consumption of a specialized sesame diet. At 4 weeksof age, birds with similar body mass (BM) were sub-groupedfor feeding behaviour measurement. Starting with the heav-ier BM, two birds of similar BM from each group at a timewere brought from the rearing house to the measurement lab-oratory. At the measurement lab, the birds were individuallyhoused in cages (24 cm (W) × 43 cm (D) × 40 cm (H)) at con-stant ambient temperature of 24 ◦C and relative humidity of46%. After a 2-day acclimation, feeding behaviour was mon-itored for the next 45 h, and those data associated with thefinal 24 h (6 am–6 am) were taken to be representative of the

P

COMPAG 2049 1–7behaviour from time-series weigh scale records, Comput. Electron.

normal behaviour of the birds and thus used in the analysis. 169

Full details are available from Xin (2001, unpublished research 170

report). For purposes of this study, sample sequences from a 171

single day of four birds were used.

teps of algorithm AL2.

ED

OF

IN PRESSCOMPAG 2049 1–7

a g r i c u l t u r e x x x ( 2 0 0 7 ) xxx–xxx

172

173

174

175

176

177

178

179

180

181

182

183

184

185

186

187

188

189

190

191

192

193

194

195

196

197

198

199

200

201

202

203

204

205

206

207

208

209

210

211

212

213

214

215

216

217

218

219

220

221

222

223

Fig. 3 – Representative high-frequency time-series weigh

224

225

226

227

228

229

230

231

232

233

234

10, and 15; Ts = 5, R = 2, 4, and 6; Ts = 10 and 30, R = 2, 3, 4, and 235

5). Additionally, the effect of the meal weight threshold (WT) 236

used to determine start of a feeding event was evaluated by 237

NC

OR

RE

CT

ARTICLE4 c o m p u t e r s a n d e l e c t r o n i c s i n

2.3. Algorithm development

Two algorithms (AL1 and AL2) were developed to utilize time-series recordings of feeder weights as the basis for assessingindividual bird meal activity. Both algorithms were designedto post-process large volumes of feeder weight recordings. AL1was designed to handle high-frequency (10 Hz) time-seriesrecordings, whereas AL2 was designed for lower frequency(1/30–1/4 Hz; or sample times Ts of 30 or 4 s). The frequencycriteria were dictated by the instrumentation systems used toacquire the data, and offered a unique opportunity to assesshow well each algorithm performed, and the importance ofsampling frequency on determining behavioural attributes. Arepresentative flow chart of the main decision steps of AL2 ispresented in Fig. 2.

Each algorithm processes a series of feeder weight readingstaken at discrete times tk, denoted by W(tk) = Wk, where indexk = 1,2,. . .n denotes sequential recordings taken at a samplerate of Ts sec. For both algorithms, the following key elementswere used:

• comparison of Wk to a threshold weight to assess whethera candidate feeding event occurred;

• determination of event start times and duration;• determination of whether each event represents feeding or

non-feeding activity.

For AL1 (Ts = 0.1 s), key features and discriminant stepsinclude:

• a stabilized baseline feed weight from an arithmetic mov-ing average (ARMA) of “R” consecutive readings is used todetermine whether the next Wk is the start of a candidatefeeding event;

• use of a forward-based “R”-pt ARMA to determine mealevent cessation;

• feeding event assessment using a 0.2 g threshold betweenstart and end weights;

• automated handling of tare, when feed was added to thesystem.

For AL2 key discriminant steps include:

• compare the sequential differences �Wk = Wk − Wk−1 to athreshold weight to assess whether a candidate event hasoccurred;

• determine candidate event duration and end time from athreshold based on Ts;

• assign candidate event activity as either feeding or non-feeding using an ARMA of time-series weights for before-and after-meal tare.

Representative weigh-scale readings were used to assess algo-rithm performance. One sample (Fig. 3) was obtained from thehigh-frequency data set (T = 0.1 s, 110 min total), and the other

U

Please cite this article in press as: Gates, R.S., Xin, H., Extracting poultryAgric. (2007), doi:10.1016/j.compag.2007.08.015

s

from a full day of recordings (Fig. 4, Ts = 4 s) in which the hen’sactivities were recorded with time-lapse video. Each algorithmwas studied to assess robustness to tuning parameters, andsampling frequency, as described below.

PR

Oscale recording (Ts = 0.1 s) used for algorithm tuning. Circlesrepresent candidate event starts from AL2.

AL1 was considered optimally tuned for discerning feedingactivity statistics from the high-frequency data, and suitablefor use in assessing dietary and environment effects on indi-vidual birds. However, a reduction in sampling frequency wasof interest to reduce storage requirements and processingtimes. Thus, 110 min of representative data sampled at 10 Hz(Fig. 3) were decimated (Ts = 1, 2, 5, 10 and 30 s) and analyzedwith different algorithm parameters to determine if similarconclusions could be drawn. Specifically, the number of sam-ples in the ARMA, R, was adjusted with sampling interval Ts

(Ts = 0.1, R = 100 and 200; Ts = 1, R = 10, 20, and 30; Ts = 2, R = 5,

COMPAG 2049 1–7behaviour from time-series weigh scale records, Comput. Electron.

Fig. 4 – Representative time-series weigh scale recording(Ts = 4 s) used for algorithm tuning, representing a full(05:00–20:00 inclusive). Circles represent candidate eventstarts from AL2.

D

INCOMPAG 2049 1–7

a g r

p238

0239

a240

(241

r242

l243

244

t245

a246

p247

t248

a249

(250

v251

p252

253

n254

f255

d256

b257

e258

p259

t260

u261

c262

e263

s264

3

3265

T266

i267

268

269

270

271

272

273

274

275

276

277

278

279

280

281

282

283

284

285

286

287

288

289

290

291

292

293

294

295

296

297

298

ARTICLEc o m p u t e r s a n d e l e c t r o n i c s i n

erforming the analysis at 0.5 g (original tuned value) and at.2 g. The effects of tuning these parameter combinations wasssessed from computed meal size (MS (g)), meal duration (MDs)), time between meals (MI: meal interval (s)) and ingestionate (IR (g min−1)). Performance of AL2 when analyzing theower frequency IBU data was also evaluated.

Parameters for AL2 that were adjusted included weighthreshold (WT (g)), end of event threshold (EET, no. samples),nd the number of points in the ARMA (ARMAnpoints). A com-arative assessment of AL2 versus AL1 was made by analyzinghe same high-frequency (and decimated subset) data (Fig. 3),nd lower sampling frequency data obtained from the IBUFig. 4). Observations of bird behaviour taken from time-lapseideo of the IBU data were used to direct the tuning of AL2arameters.

AL2 was developed with a different set of criteria than AL1,amely to identify both feeding and non-feeding activities

rom lower sampling frequency data. Thus statistics on “can-idate events”, i.e. activity at a feeder that may or may note feeding, were gathered. To discriminate feeding events, anvent key code (0 or 1) was assigned for each event by com-aring feed disappearance for each event to the meal weighthreshold. Total feed consumed was obtained from a dot prod-ct between the event key code array and individual feedhanges for each candidate event. The sum of entries in thevent key code yields the number of meals, and other statisticsuch of mean MS, MD and IR were computed.

. Results and discussion

UN

CO

RR

EC

TE

Please cite this article in press as: Gates, R.S., Xin, H., Extracting poultryAgric. (2007), doi:10.1016/j.compag.2007.08.015

.1. Algorithm AL1

he algorithm was tested on the representative feed sequencen Fig. 3, and evaluated for different values of sampling interval

Table 1 – Hourly summaries of bird behaviour from video recor

Time (hh:mm) Number of feedingevents

Number of drinkievents

5:00 5 06:00 3 47:00 7 48:00 2 29:00 7 410:00 4 311:00 5 212:00 8 2

5:00–12:00 Total (%) 41 21

14:00 7 515:00 1 016:00 8 617:00 6 618:00 5 319:00 10 820:00 5 4

14:00–20:00 Totals (%) 42 32

PR

OO

F

PRESSi c u l t u r e x x x ( 2 0 0 7 ) xxx–xxx 5

Ts, number of points in the ARMA, R, and two different weightthresholds. Complete results are available in Table 1 of Gatesand Xin (2001) and not repeated here. The actual change infeed mass during this 110 min period was 5.3 g, and the tunedalgorithm from the full data set (Ts = 0.1 s, R = 100) yielded thefollowing baseline information: four meals totalling 5.2 g, withmean meal statistics of meal size (MS) = 1.3 g, meal duration(MD) = 89 s, meal interval (MI) = 1088 s, and meal ingestion rate(IR) = 0.9 g min−1.

The effect of doubling R was to increase mean MD by 6 s,and did not affect other statistics. For Ts = 1 s and R = 20, thebaseline case was closely mimicked, whereas for R = 10 (i.e. a10 s ARMA) IR was over-predicted. For Ts = 2 s, R = 15 producedresults similar to the baseline (MD was 8 s greater) and bothR = 5 and R = 10 matched actual feed used. At Ts = 5 s, a 6-ptARMA produced results most similar to the baseline exceptthat it missed one meal.

The effect of using a smaller meal weight threshold WT(0.2 g vs. 0.5 g) was as follows. For shorter Ts, the shorter ARMAgave similar MS but one extra meal and thus shorter MD andMI, and greater IR. This trend of counting additional meals wasalso noted at greater Ts, and combinations in which Ts × R = 30gave results most similar to baseline whereas Ts × R = 10 over-predicted total feed used. Results for (Ts = 10, R = 4) also closelymatched baseline results. Surprisingly, Ts = 30, R = 2 predictedtotal feed, MS and MD quite well.

The opportunity for 10–20-fold reduction in sampling fre-quency when using AL1 thus appears reasonable. Attemptingto improve sensitivity by decreasing the threshold to 0.2 g cre-ates erroneous counts of additional meals noted and changesmeal statistics. This latter point underlines the importance

COMPAG 2049 1–7behaviour from time-series weigh scale records, Comput. Electron.

of devising a set of definitions for what constitutes a meal. 299

Once defined, AL1 with reduced sampling interval time-series 300

data shows great promise in automating feeding activity 301

analysis.

dings (hen 4, thermoneutral conditions)

ng Total time (s) at Feed consumed(g)

Feeder Waterer Other

160 0 3441 3350 221 3036 5.8550 191 2857 6.5498 174 2927 7.4950 248 2403 9.6528 208 2863 3.5572 114 2911 6507 153 2936 3.5

4115 (14.3) 1309 (4.5) 23374 (81.2) 37.9

762 198 2645 7.957 0 3543 0.8

598 186 2821 7652 278 2672 6.9721 164 2715 9.8

1425 446 1573 16.5620 129 2850 7.9

4835 (19.2) 1401 (5.6) 18819 (74.7) 56.8

ED

ARTICLE IN PRESSCOMPAG 2049 1–7

6 c o m p u t e r s a n d e l e c t r o n i c s i n a g r i c u l t u r e x x x ( 2 0 0 7 ) xxx–xxx

Table 2 – Summary of results using AL2 on the data in Table 1, with parameters: WT = 0.5 g, EET = 12 and ARMAnpoints = 3

Time period Total event time Total feeding time Mean meal duration Mean meal interval Total feed consumed (g)

5:00–12:00(s) 4436 3724 186 767 39.5% 17.6 14.8Error (%) −9.5 4.2

14:00–20:00(s) 5512 4760 176 621 53.5% 21.9 18.9Error (%) −1.6 −5.8

Totals(s) 9948 8484 362 1388 93.0

302

303

304

305

306

307

308

309

310

311

312

313

314

315

316

317

318

319

320

321

322

323

324

325

326

327

328

329

330

331

332

333

334

335

336

337

338

339

340

341

342

343

344

345

346

347

348

349

350

351

352

353

354

355

356

357

358

359

360

361

362

363

364

365

366

367

368

369

370

371

372

373

374

375

376

NC

OR

RE

CT

% 19.7 16.8Error (%) −5.2

3.2. Algorithm AL2

The algorithm was tested on the representative feedingsequence in Fig. 3, and evaluated for different values of sam-pling interval Ts, meal weight threshold WT, end of eventthreshold EET, and number of points in the ARMA, ARMAn-points. Complete results are available in Table 1 of Gates andXin (2001) and not repeated here.

For the full data set (Ts = 0.1 s), AL2 mimicked AL1 pre-dictions closely (with parameters: WT = 0.5 g; EET = 100, 200or 300; and ARMAnpoints = 120). Both algorithms give simi-lar meal statistics, however, AL2 estimated meal interval to be400 s, whereas AL1 estimated it to be 1088 s (and it cannot bestated which is more accurate).

A 10-fold increase in sampling interval (Ts = 1 s), andreduced weight threshold (WT = 0.2 g) resulted in over-estimated number of meals and associated feeding statistics.However, increasing WT to 0.5 g provided good estimates ofmeal events, and with EET = 8 and 12, the results from AL1were bracketed. Further increasing EET to 30 appeared to pro-vide the best match with AL1. The number of non-meal eventspredicted were 31, 26 or 20 for these values of EET, respec-tively. For sampling interval, Ts = 2 s, total feed consumed wasunderestimated (EET = 8, 12, 30), but total feeding time wasreasonably matched. For Ts = 5 s, a combination of WT = 0.5and EET = 4 was best, although total feed consumed and feed-ing time were slightly over-estimated. At Ts = 10 s, consumedfeed was slightly under-predicted and total feeding time wasover-predicted, with EET = 4 or 8 giving best results. At Ts = 30 s,predictions were not very consistent. From these results, itappears that using a 30 s sampling interval with AL2 is notadvisable, but Ts = 1, 2, 5 or 10 s were practical with appropriatetuning of WT and EET.

3.3. Comparison to video data

Sample summary behavioural data taken from time-lapsedvideo recordings are presented in Table 1. The data includenumber of feeding and drinking events, and time at feeder,

U

Please cite this article in press as: Gates, R.S., Xin, H., Extracting poultryAgric. (2007), doi:10.1016/j.compag.2007.08.015

waterer, and other. The data are summarized by hour,and by an 8 h morning and 7 h evening period (5:00–12:00and 14:00–20:00 inclusive, respectively) covering the day-light hours. Weigh scale data were simultaneously recorded

PR

OO

F −1.8

with Ts = 4 s. A plot of the time-series of weights is givenin Fig. 4.

For the first time period, AL1 with R = 5 and a weight thresh-old of 0.2 or 0.5 g, detected respective values of total feedconsumed of 41.8 and 41.7 g, and 3180 and 2924 s spent eat-ing (12.6% and 11.6%, respectively vs. 14.3% observed). Duringthe second time period, total feeding time was 4636 s and4240 s (18.4 and 16.8% of total time) for the two values of WT,respectively, compared to 18.9% observed; and predicted totalfeed consumed was 55.8 g, compared with 56.8 observed (2%under-estimate).

Table 2 presents summary results from applying AL2 tothe sample day’s data. Selected parameters for AL2 wereWT = 0.5 g, EET = 12 and ARMAnpoints = 3. Total period feedconsumption was 93 g compared with 94.7 g obtained by sub-traction from the stored records. Predicted time at feeder(Total Event Time) was 17.6% versus 14.3% noted from videorecordings, and 21.9% versus 19.2%, for the two time periods,respectively. MD and MI averaged 362 and 1388 s for the day,with 20 and 27 meals noted for period 1 and period 2, respec-tively. For the entire period, 85% of time at feeder involvedeating, and the remainder was classified as stereotyped peck-ing. This statistic could not be checked against the videorecordings, but demonstrates the ability of AL2 to discern notonly meal size and duration, but also non-meal activity at thefeeder.

3.4. Other effects

A low frequency band pass filter (pass band 1e−2 to 5e−0 Hz)was applied to the high-frequency data, and the resultant fil-tered data were subjected to analysis with both algorithms.AL1 provided nearly identical results with the filtered data atTs = 0.1 s, if WT was held at 0.5 g; however, six meals and atotal of 5.8 g feed consumption were predicted if meal weightthreshold WT was reduced to 0.2 g. AL2 over predicted numberof meals and total feed consumed.

COMPAG 2049 1–7behaviour from time-series weigh scale records, Comput. Electron.

4. Summary and conclusions

Applications of results from individual bird activity at the 377

feeding station include development of a frequency of occur- 378

D

INCOMPAG 2049 1–7

a g r

r379

o380

d381

p382

b383

a384

e385

a386

387

s388

t389

p390

r391

i392

o393

u394

a395

396

m397

l398

d399

i400

i401

u402

n403

404

•405

406

407

•408

409

410

•411

412

413

•414

415

416

417

UQ2

P418

A

F419

H

420

421

422

423

424

r425

426

427

428

429

430

431

432

433

434

435

436

437

438

439

440

441

442

443

444

445

446

447

448

449

450

451

452

453

454

455

456

457

458

459

460

461

462

463

464

465

466

467

drinking patterns of broilers subjected to different feeding 468

RR

EC

TE

ARTICLEc o m p u t e r s a n d e l e c t r o n i c s i n

ence estimate, or histogram, for statistics including numberf meals per unit time, MS, MD, MI and IR. These statistics,erived from autonomous real-time or stored and post-rocessed data, can provide an objective basis for evaluatingird response to environmental or management stimuli. Thelgorithms presented in this work have been successfullymployed in a series of subsequent research trials (Persyn etl., 2003, 2004; Cook and Xin, 2004).

With signal processing such as outlined in this paper, mealize, frequency of occurrence, and a measure of bird stereo-yped pecking at the feed station can be obtained. While theroposed technique cannot be made to distinguish betweenesting, preening or other activities away from the feeder,t can be used to assess impact of environmental stressorsn feeding behaviour, and to determine how dietary manip-lation may be utilized to counteract deleterious effects ofdverse environmental conditions.

Additional signal processing is recommended for improve-ent of predictions. For example, a properly configured

ow-pass filter applied to the data may enhance meal sizeeterminations without degrading meal duration and interval

nformation. Use of a self-tuning black box linear model, cal-brated and validated against each bird’s behaviour, could besed for real-time assessment of dynamic response to exter-al stimuli.

In conclusion:

Both algorithms, AL1 and AL2, could be tuned to providesimilar predictions, thus a wide range in data sampling fre-quency could be analyzed.AL1 was developed for 10 Hz time-series recordings(Ts = 0.1 s); however, it was found to robustly determine mealsize, duration and interval information up to Ts = 1–30 s.AL2 was developed for lower sampling frequency data, yetreasonably matched AL1 results for data taken at Ts = 0.1and 1 s.Both algorithms were capable of predicting time at feederwith good agreement with observed video recordings. Theyprovide the additional benefit of discrimination betweeneating at the feeder, versus stereotyped pecking.

ncited reference

ersyn et al. (2002).

UN

CO

Please cite this article in press as: Gates, R.S., Xin, H., Extracting poultryAgric. (2007), doi:10.1016/j.compag.2007.08.015

cknowledgements

unding for the research was partially provided by Ajinomotoeartland Inc., and a grant from USDA NRI Animal Health and

PR

OO

F

PRESSi c u l t u r e x x x ( 2 0 0 7 ) xxx–xxx 7

Well-Being Program, and is acknowledged with gratitude. Thispaper no. 01-05-111 of the Kentucky Agricultural ExperimentStation, reports results of an investigation by the Kentucky andIowa Agricultural Experiment Stations, and is published withthe approval of the Directors.

e f e r e n c e s

Abeyesinghe, S.M., 2000. Aversion of the domestic fowl toconcurrent stressors: methodology. PhD Thesis. University ofBristol, UK.

Cook, R.N., Xin, H., 2004. Effects of cage stocking density onfeeding behaviours of group-housed laying hens. In: ASAEInternational Meeting, Ottawa, ON, Canada (ASAE Paper no.044004).

Gates, R.S., Xin, H., 2001. Comparative analysis of measurementtechniques of feeding and drinking behaviour of individualpoultry subjected to warm environmental condition. In: ASAEInternational Meeting, Sacramento, CA, USA (ASAE Paper no.014033).

Gates, R.S., Turner, L.W., Chi, H., Usry, J., 1995. Automatedweighing of group-housed growing–finishing swine. Trans.ASAE 38 (5), 1479–1486.

Naas, I.A., Pereira, D.F., Curto, F.P.F., Behrens, F.H., Carvalho, J.C.C.,Amendola, M., Mantovani, E.C., 2000. Proceedings of The XIVMemorial CIGR World Congress. Determining the BroilerFemale Breeder Behaviour Using Telemetry, vol. 1, CIGR,Tsukuba, Japan, pp. 1014–1018.

Persyn, K.E., Xin, H., Gates, R.S., 2002. Effects of beak trimming onpoultry ingestion behaviour. In: ASAE International Meeting,Chicago, IL, USA (ASAE Paper no. 024070).

Persyn, K.E., Xin, H., Ikeguchi, A., Gates, R.S., 2003. Feedingbehaviours and pecking force of chicks with or without beaktrimming. In: ASAE International Meeting, Las Vegas, NV, USA(ASAE Paper no. 034005).

Persyn, K.E., Xin, H., Nettleton, D., Ikeguchi, A., Gates, R.S., 2004.Feeding behaviour of laying hens with or without beaktrimming. Trans. ASAE 47 (2), 591–596.

Puma, M.C., Xin, H., Gates, R.S., Burnham, D.J., 2001. Aninstrumentation system for measuring feeding and drinkingbehaviour of individual poultry. Appl. Eng. Agric. 17 (3),365–374.

Wathes, C.M., Abeyesinghe, S.M., Frost, A.R., 2001. Environmentaldesign and management for livestock in the 21st century:resolving conflicts by integrated solutions. In: Stowell, R.R.,Bucklin, R., Bottcher, R.W. (Eds.), Proceedings of LivestockEnvironment VI. 21–23 May. ASAE Publications, St. Joseph, MI,pp. 5–14.

Xin, H., Berry, I.L., Barton, T.L., Tabler, G.T., 1993. Feeding and

COMPAG 2049 1–7behaviour from time-series weigh scale records, Comput. Electron.

and lighting programs. Appl. Poultry Res. 2 (4), 365–372. 469

Xin, H., Gates, R.S., Puma, M.C., Ahn, D.U., 2002. Drinking water 470

temperature effects on laying hens subjected to warm cyclic 471

environments. Poultry Sci. 81, 608–617. 472