Background and hypotheses – PigIT, University of Copenhagen

PigIT > About PigIT > Background and hypotheses

Background and hypotheses of the project

In the last ten years, noticeable progress in terms of productivity has been reported for sow herds. This has resulted in increased potential for the value chain. However, the part of the value chain coming from growing pigs is currently not fully exploited. This is due to an unsatisfactory level of both welfare and productivity of growing pigs, and to the increasing export of live piglets observed in recent years.

The major welfare problems affecting the productivity of growing and finishing pigs are intestinal (i.e. diarrhea, affecting mostly weaners) and respiratory (affecting mostly finishers) diseases. Other major welfare issues are tail biting and undesired excretory behavior with sudden appearance of filthy floors. The causes of both tail biting and undesired excretory behavior are currently unclear, which makes it difficult to take action to tackle and prevent these welfare issues. Tail docking, which is a painful intervention to prevent tail biting, is only allowed in herds with severe tail biting problems. However, tail biting is currently so common, that tail docking is almost routine practice. Indeed, a better understanding of the causes of tail biting is necessary to avoid both its occurrence and also the practice of tail docking.

It is the hypothesis that a systematic placement of cheap sensors in the production pens combined with methodological developments to integrate the information from these sensors will improve the production process and thus add significant value to investment in the sensor technology. The potential benefit will be seen in productivity as well as in the welfare of the animals in the systems.

Simple sensors are already installed on many pig farms (e.g. temperature, water and feed). Typically, the numbers and positions of the sensors are not directly related to the position and design of the production pens, and it is not possible to e.g. monitor the production in each pen consistently. However, addition of a few sensors plus a systematic approach to the positioning of the sensor promise to add new dimension to the use of the information from the sensors, both with respect to how different parts of production systems function, which types of problems that may be detected, and how the information may be used for a dynamic fine tuning of the production process.

Until now, computer based production monitoring in growing pigs has mainly focused on growth, consumption of water and feed, and mortality. However, in the daily management pig producers rely on additional observations related to animal behavior, e.g. in relation to climate regulations, and early intervention in case of economic and welfare related problems, such as disease and tail biting. These behavioral observations also serve as early indicators in case of feed and growth related problems. Several research projects described in the literature have tried to develop methods for automatic monitoring in pig production (examples include a group in Helsinki working with detection of farrowing [24] and a group in Leuven working with detection of respiratory diseases by cough sound analysis [8]). Also the groups at UCPH LAS and AU ANIS have worked intensively in this field based on techniques such as state space models [4, 5, 6, 21] and Hidden Markov models [2, 3]. Data sources have been vision technology [3, 11, 27], sensors positioned on individual animals (tri-axial accelerometers [4, 5, 6, 20], temperature sensors [1]) and sensors monitoring the environment, i.e. water [2, 19] and climate sensors. Statistical methods based on behavioural theories have been developed, e.g. to describe the choice and lying behaviour of the pigs [7, 10], and the timing of different motivational states of sows. The previous efforts in the two groups have been concentrated on monitoring sows, but the ambition of this alliance is to adapt their methodology to growing pigs and to go a step further and integrate the monitoring methods in active decision support and regulation. This will allow better integration of different information sources and from different parts of the production system. 

The research groups at UCPH LAS and AU ANIS have already contributed significantly to the development of efficient algorithms for optimization of decision problems in animal production [12, 13, 15, 21, 23].  Some of the techniques developed have been applied world-wide and are now internationally considered as standard tools. This is particularly the case with the MLHMP, Multi Level Hierarchical Markov Processes, for replacement models [15].  In this context the most important aspect is that it allows an easy specification of a wide range of different decisions, which can be specified in the specific production process, and thus can be used for specification of the interventions based on the automatic monitoring. Nevertheless, developments are needed to further improve the capabilities of the modelling framework in order to address problems of increasing biological complexity. It is expected that improved data structures and integration with graphical models (e.g. Bayesian networks or Limited Memory Influence Diagrams [17]) can contribute to solve the issue. Again, previous efforts have mainly addresses sow-related decision problems, but in the projects of this alliance the methods will be adapted to growing pigs.