A system for automated acquisition of the mean pig weight from a group of growing-finishing animals was developed and tested. Each time an animal entered a crate to obtain a drink of water, a time series of weights was recorded. A state machine representation was implemented to capture weight events and to put time series weight recordings into a queue. Three algorithms (mean, histogram, and median) were devised and evaluated to extract animal weight from the time series recording. Four tests were performed—short-term tests included test 1 with 12 animals of 64.5 to 96.4 kg initial weight; test 2 with 8 animals of 70.9 to 75 kg; test 3 with 3 individual animals of 45, 68, and 93 kg; and test 4 was for a full growth period with 10 animals. The first three experiments were used to develop and refine the system, and test 4 was used to assess the long-term performance and feasibility. Comparisons between estimated weight and “spot checks” with a mechanical scale were typically within about ±1% over test 4. The histogram algorithm was found to be the best for estimating average weight. Accuracy was not affected by variation in animal weight (tests 1 and 2); however, different length data windows affected accuracy and a 4-h window was best. A time series queue of at most 150 s of data was found to be adequate and the histogram algorithm was successful with as few as 10 data points. Locating the sole water supply within the weigh crate did not affect growth rate compared to a control group. A change rate (CR) index was devised to compare daily activity and to flag pen health problems
Read full abstract