The Sentinel Bait Station: an automated, intelligent design pest animal baiting system
G. Charlton
A * , G. Falzon A B , A. Shepley A , P. J. S. Fleming C D , G. Ballard D E and P. D. Meek D F
A School of Science and Technology, University of New England, Armidale, NSW 2351, Australia.
B College of Science and Engineering, Flinders University, Adelaide, SA 5001, Australia.
C Vertebrate Pest Research Unit, NSW Department of Primary Industries, Orange Agricultural Institute, 1447 Forest Road, Orange, NSW 2800, Australia.
D School of Environmental and Rural Science, University of New England, Armidale, NSW 2351, Australia.
E Vertebrate Pest Research Unit, NSW Department of Primary Industries, Building C02, University of New England, Armidale, NSW 2351, Australia.
F Vertebrate Pest Research Unit, NSW Department of Primary Industries, Coffs Harbour, NSW 2450, Australia.
Abstract
Ground baiting is a strategic method for reducing vertebrate pest populations. Best practice involves maximising bait availability to the target species, although sustaining this availability is resource intensive because baits need to be replaced each time they are taken. This study focused on improving pest population management through the novel baiting technique outlined in this manuscript, although there is potential use across other species and applications (e.g. disease management).
To develop and test an automated, intelligent, and semi-permanent, multi-bait dispenser that detects target species before distributing baits and provides another bait when a target species revisits the site.
We designed and field tested the Sentinel Bait Station, which comprises a camera trap with in-built species-recognition capacity, wireless communication and a dispenser with the capacity for five baits. A proof-of-concept prototype was developed and validated via laboratory simulation with images collected by the camera. The prototype was then evaluated in the field under real-world conditions with wild-living canids, using non-toxic baits.
Field testing achieved 19 automatically offered baits with seven bait removals by canids. The underlying image recognition algorithm yielded an accuracy of 90%, precision of 83%, sensitivity of 68% and a specificity of 96% throughout field testing. The response time of the system, from the point of motion detection (within 6–10 m and the field-of-view of the camera) to a bait being offered to a target species, was 9.81 ± 2.63 s.
The Sentinel Bait Station was able to distinguish target species from non-target species. Consequently, baits were successfully deployed to target species and withheld from non-target species. Therefore, this proof-of-concept device is able to successfully provide baits to successive targets from secure on-board storage, thereby overcoming the need for daily bait replacement.
The proof-of-concept Sentinel Bait Station design, together with the findings and observations from field trials, confirmed the system can deliver multiple baits and increase the specificity in which baits are presented to the target species using artificial intelligence. With further refinement and operational field trials, this device will provide another tool for practitioners to utilise in pest management programs.
Keywords: artificial intelligence, baiting, invasive species, machine vision, pest control, pest management, recognition, 1080.
References
Allen LR, Fleming PJS, Thompson JA, Strong K (1989) Effect of presentation on the attractiveness and Palatability to wild dogs and other wildlife of 2 unpoisoned wild-dog bait types. Wildlife Research 16, 593-598.
| Crossref | Google Scholar |
Ballard G, Fleming PJS, Meek PD, Doak S (2020) Aerial baiting and wild dog mortality in south-eastern Australia. Wildlife Research 47, 99-105.
| Crossref | Google Scholar |
Busana F, Gigliotti F, Marks CA (1998) Modified M-44 cyanide ejector for the baiting of red foxes (Vulpes vulpes). Wildlife Research 25, 209-215.
| Crossref | Google Scholar |
Corva DM, Semianiw NI, Eichholtzer AC, Adams SD, Mahmud MAP, Gaur K, Pestell AJL, Driscoll DA, Kouzani AZ (2022) A smart camera trap for detection of endotherms and ectotherms. Sensors 22, 4094.
| Crossref | Google Scholar |
Dexter N, Meek P, Moore S, Hudson M, Richardson H (2007) Population responses of small and medium sized mammals to fox control at Jervis Bay, Southeastern Australia. Pacific Conservation Biology 13, 283-292.
| Crossref | Google Scholar |
Falzon G, Lawson C, Cheung K-W, Vernes K, Ballard GA, Fleming PJS, Glen AS, Milne H, Mather-Zardain A, Meek PD (2020) ClassifyMe: a field-scouting software for the identification of wildlife in camera trap images. Animals 10, 58.
| Crossref | Google Scholar |
Fleming PJS (1996) Ground-placed baits for the control of wild dogs: evaluation of a replacement-baiting strategy in north-eastern New South Wales. Wildlife Research 23, 729-740.
| Crossref | Google Scholar |
Fleming PJS (1997) Uptake of baits by red foxes (Vulpes vulpes): implications for rabies contingency planning in Australia. Wildlife Research 24, 335-346.
| Crossref | Google Scholar |
Fleming PJS, Allen LR, Berghout MJ, Meek PD, Pavlov PM, Stevens P, Strong K, Thompson JA, Thomson PC (1998) The performance of wild-canid traps in Australia: efficiency, selectivity and trap-related injuries. Wildlife Research 25, 327-338.
| Crossref | Google Scholar |
Fleming PJS, Ballard G, Reid NCH, Tracey JP (2017) Invasive species and their impacts on agri-ecosystems: issues and solutions for restoring ecosystem processes. The Rangeland Journal 39, 523-535.
| Crossref | Google Scholar |
Howard AG, Zhu M, Chen B, Kalenichenko D, Wang W, Weyand T, Andreetto M, Adam H (2017) Mobilenets: efficient convolutional neural networks for mobile vision applications.
| Crossref | Google Scholar |
Marks CA, Wilson R (2005) Predicting mammalian target-specificity of the M-44 ejector in south-eastern Australia. Wildlife Research 32, 151-156.
| Crossref | Google Scholar |
Marks CA, Busana F, Gigliotti F (1999) Assessment of the M-44 ejector for the delivery of 1080 for red fox (Vulpes vulpes) control. Wildlife Research 26, 101-109.
| Crossref | Google Scholar |
Meek PD, Brown SC (2017) It’s a dog eat dog world: observations of dingo (Canis familiaris) cannibalism. Australian Mammalogy 39, 92-94.
| Crossref | Google Scholar |
Meek PD, Jenkins DJ, Morris B, Ardler AJ, Hawksby RJ (1995) Use of two humane leg-hold traps for catching pest species. Wildlife Research 22, 733-739.
| Crossref | Google Scholar |
Meek P, Ballard G, Fleming P, Falzon G (2016) Are we getting the full picture? Animal responses to camera traps and implications for predator studies. Ecology and Evolution 6, 3216-3225.
| Crossref | Google Scholar |
Meek PD, Ballard G, Falzon G, Williamson J, Milne H, Farrell R, Stover J, Mather-Zardain AT, Bishop JC, Cheung EK-W, Lawson CK, Munezero AM, Schneider D, Johnston BE, Kiani E, Shahinfar S, Sadgrove EJ, Fleming PJS (2020) Camera trapping technology and related advances: Into the new millennium. Australian Zoologist 40, 392-403.
| Crossref | Google Scholar |
Meek PD, Ballard GA, Mifsud G, Fleming PJS (2022) Foothold trapping: Australia’s history, present and a future pathway for humane use. In ‘Mammal trapping: wildlife management, animal welfare & international standards. Vol. 5’. (Ed. G Proulx) pp. 81–95. (Alpha Wildlife Publications: Alberta, Canada)
Moseby KE, McGregor H, Read JL (2020) Effectiveness of the Felixer grooming trap for the control of feral cats: a field trial in arid South Australia. Wildlife Research 47, 599-609.
| Crossref | Google Scholar |
Newsome A (1990) The control of vertebrate pests by vertebrate predators. Trends in Ecology & Evolution 5, 187-191.
| Crossref | Google Scholar |
Ross R, Parsons L, Thai BS, Hall R, Kaushik M (2020) An IoT smart rodent bait station system utilizing computer vision. Sensors 20, 4670.
| Crossref | Google Scholar |
Sadgrove EJ, Falzon G, Miron D, Lamb DW (2018) Real-time object detection in agricultural/remote environments using the multiple-expert colour feature extreme learning machine (MEC-ELM). Computers in Industry 98, 183-191.
| Crossref | Google Scholar |
Shahinfar S, Meek P, Falzon G (2020) “How many images do I need?” Understanding how sample size per class affects deep learning model performance metrics for balanced designs in autonomous wildlife monitoring. Ecological Informatics 57, 101085.
| Crossref | Google Scholar |
Shepley A, Falzon G, Kwan P (2020) Confluence: a robust non-IoU alternative to non-maxima suppression in object detection.
| Crossref | Google Scholar |
Shepley A, Falzon G, Lawson C, Meek P, Kwan P (2021a) U-infuse: democratization of customizable deep learning for object detection. Sensors 21, 2611.
| Crossref | Google Scholar |
Shepley A, Falzon G, Meek P, Kwan P (2021b) Automated location invariant animal detection in camera trap images using publicly available data sources. Ecology and Evolution 11, 4494-4506.
| Crossref | Google Scholar |
