A Monitoring System for Laying Hens That Uses a Detection Sensor Based on Infrared Technology and Image Pattern Recognition

Abstract

In Italy, organic egg production farms use free-range housing systems with a big outdoor area and a flock of no more than 500 hens. With additional devices and/or farming procedures, the whole flock could be forced to stay in the outdoor area for a limited time of the day. As a consequence, ozone treatments of housing areas could be performed in order to reduce the levels of atmospheric ammonia and bacterial load without risks, due by its toxicity, both for hens and workers. However, an automatic monitoring system, and a sensor able to detect the presence of animals, would be necessary. For this purpose, a first sensor was developed but some limits, related to the time necessary to detect a hen, were observed. In this study, significant improvements, for this sensor, are proposed. They were reached by an image pattern recognition technique that was applied to thermografic images acquired from the housing system. An experimental group of seven laying hens was selected for the tests, carried out for three weeks. The first week was used to set-up the sensor. Different templates, to use for the pattern recognition, were studied and different floor temperature shifts were investigated. At the end of these evaluations, a template of elliptical shape, and sizes of 135 × 63 pixels, was chosen. Furthermore, a temperature shift of one degree was selected to calculate, for each image, a color background threshold to apply in the following field tests. Obtained results showed an improvement of the sensor detection accuracy that reached values of sensitivity and specificity of 95.1% and 98.7%. In addition, the range of time necessary to detect a hen, or classify a case, was reduced at two seconds. This result could allow the sensor to control a bigger area of the housing system. Thus, the resulting monitoring system could allow to perform the sanitary treatments without risks both for animals and humans.

Keywords: infrared sensors; laying hens; organic egg production systems; ozone; patterns matching.

Figure 1

The picture shows the set-up of the experimental housing system and developed detection sensor. The water dispenser, that is represented with the symbol “W”, and the other main components of the housing system are reported. In the closed room, drawn in red, are shown: the detection sensor, mounted on the ceiling of the room, and its field of view (FOV) highlighted on the floor of the room. Finally, the dimensions of the housing system, of the sensor position and of its FOV, are also reported.

Figure 2

Components and connections of the experimental monitoring system. In the figure, it is also reported a commercial web-cam. This component was added to the monitoring system in order to collect, during the experiments, images of hens reared in the housing system. These images were used to evaluate the performance achieved by the detection sensor under study.

Figure 3

Flow diagram of the software application of the monitoring system.

Figure 4

Flow diagram of the elaborations performed by the detection subroutine. The procedures of image pre-processing, pattern recognition and post-processing of acquired data are explained in detail. Each blue rectangle describes an elaboration performed by the subroutine which is linked to the following step of elaboration through a blue arrow. For some specific elaborations, with a grey arrow is highlighted an input while with a red arrow is shown an output.

Figure 5

Pictures (A–C) are thermografic images of a hen acquired by the detection sensor during field tests. Pictures (D–F) are the results of the elaborations performed by the sensor considering as input the pictures (A–C). Picture (F) is an example of positive pattern recognition while pictures (D,E) are examples of negative pattern recognitions. Nevertheless, in picture (E), the software application counted 2089 colored pixels while in picture (D), the total amount of colored pixels was 1525. Therefore, at the end of the elaborations performed by the software application, the case of picture (E) was classified as positive while the case of picture (D) was confirmed as negative. Thus, in picture (D) is shown a false negative case while in pictures (E,F) are reported two examples of cases correctly classified as positive by the detection sensor developed.