Body weight measurement of pigs is essential for monitoring performance, welfare and overall production value. Substantial economic and production losses occur when pigs are marketed too heavy or too light.  Direct weight measurement provides the most accurate results; however, it is time-consuming, and potentially leads to stress on the animals. Alternatively, finishing pigs body weights are determined by subjective visual evaluations by experienced caretakers. This method, however, lacks consistency and accuracy. 

Another possibility for collecting weights on farms is commercial digital camera systems that automatically estimate pigs' weight. Producers can consider these systems as an alternative method for determining and monitoring pigs' body weight. 

There are two types of digital camera system for pigs' weight estimation available commercially: handheld systems and ceiling/wall mounted systems. Both types of system use 3D cameras to record top-view images and videos of pigs. All systems come with computer software to determine estimated weights based on the recorded images and videos. 

Ceiling/wall mounted systems monitors pigs' weights over time, while handheld systems estimate pigs' weight on the spot and do not provide pigs' over time performance. Examples of a ceiling/wall mounted system and handheld system are shown in Figure 1. 

Ceiling/wall mounted systems are more common than handheld systems. The cost of these systems varies, with handheld systems on the cheaper end of the spectrum, mainly due to the cost of installing the mounted systems. A common handheld system, which is more lightweight, includes a camera scanner, a tablet for data processing and displaying results. Common ceiling/wall mounted system includes a camera scanner that can connect remotely to a computer system for storing and processing recorded images and videos. Users will have access to this remote computer system to see the result of pigs' performance and weight estimation. It is required that users have a local computer system (PC or laptop) to connect to the remote computer system. 

Table 1 includes the information about some commercially available systems. We do not endorse these systems over other systems that are not included; these are purely examples for consideration. In addition, the quality and accuracy of these systems have not been validated by either consumers or researchers. 

Multiples studies were conducted in our research laboratory at North Carolina State University to validate new weight measurement technologies and determine their usefulness on swine farms. Accuracy of three methods were evaluated: human observation, a walk-across platform scale (CIMA; Correggio, Italy), and PigVision mounted cameras (Asimetrix Inc.; Durham, NC). Additionally, the application of a novel, handheld, portable RGB and stereo vision system for estimating pig body weight rapidly using images from various angles was also evaluated.   

In the first study, a trained individual selected pigs estimated to be market weight at two finishing sites. Site one had 468 pigs and an the individual recorded an accuracy of 84.4% when selected pigs that were of appropriate market weight. Site two had 522 pigs and an the individual recorded a 82.5% accuracy. Accuracy was measured by whether the pig was marked correctly in the market weight range. In addition, both image data were collected using an Intel RealSense camera from these pigs. A 3D generative model was used. The objective of this study was to validate new weight measurement technologies and determine their usefulness on swine farms. Accuracy of three methods were evaluated: human observation, a walk-across platform scale (CIMA; Correggio, Italy), and PigVision mounted cameras (Asimetrix Inc.; Durham, NC). Additionally, the application of a novel, handheld, portable RGB and stereo vision system for estimating pig body weight rapidly using images from various angles was also evaluated. 

In the first study, a trained individual selected pigs estimated to be market weight at two sites. Site one had 468 pigs and an accuracy of 84.4%, site two had 522 pigs and an 82.5% accuracy. Accuracy was measured by whether the pig was marked correctly in the market weight range. In addition, image data were collected using an Intel RealSense camera from these pigs. A 3D generative model was then used to identify latent features from each image selected, which were then used to estimate pig weights. Three regression models were examined and compared to two baseline models (median prediction and linear regression between heart girth and weight) to determine how well the models could predict pig weights. Out of the three models that were generated, the model which performed the best gave an R^2 of 0.4167 of pig weights.  This still fell short when compared to the linear regression (heart girth and weight) baseline model which provided an R^2 = 0.8621 of pig weights. More work is necessary with the image data to filter images which may not capture the whole pig as well as collecting images of pigs with a wider weight range which we anticipate would lead to a better model of pig weights.  

A 16-week study was then conducted to determine PigVision camera accuracy over time from placement to market. Cameras were mounted above 12 pens. Weights were validated every two weeks. PigVision accuracy was measured by the difference in the recorded weight from the device and the calibrated scale weight. The accuracy for pigs that weighed 32.7 kilograms (87.7%) was lower (P < 0.05) than the accuracy for pigs that weighed 117.5 kg (97.6%) or 125.7 kg (96.6%). The overall accuracy from placement to market was 94.1%. A final study at market compared visual evaluation, the walk-across scale and PigVision. A total of 91 pigs were weighed with each method. The accuracy for the walk-across scale was 98.2%. The walk-across scale did not register a weight for six pigs due to them stepping off the side of the scale or running across the scale too quickly. Final accuracies were 88.2% for visual evaluation, and 96.6% for PigVision.  

Human observation is the chosen method in many operations today yet offers the lowest accuracy. The walk-across scale is easy to operate but requires tactical animal movement. PigVision is the least arduous option, provides constant data, but does require maintenance. Ultimately, producers need a simple method that requires minimal human input, minimizes stress on the pig and is accurate enough to reduce sort loss.  While still not universally accepted, commercial digital camera systems continue to improve and may become a viable option to help reduce labor on farms. 

Source: Anh Nguyen and Jonathan Holt, who is solely responsible for the information provided, and wholly owns the information. Informa Business Media and all its subsidiaries are not responsible for any of the content contained in this information asset. The opinions of this writer are not necessarily those of Farm Progress/Informa.