16:30 〜 16:45
[4-1600-D-03] Handy Type Pig Weight Estimation System Based on Random Forest Algorithm
キーワード：Xtion-2 Device, Laser Slit, Region Growing, Random Forest, Pig Weight Estimation
In every pig farm, manual pig weight measurement takes time and needs many labors. Generally, load cell is used in pig farms to measure the pig weight. That way of measurement is hard to guide pigs to the weighting machine. The most problem is having vibration when the pig is on the load cell. It causes the inaccuracy result in measuring the pig weight and takes at least 20 seconds to get the stable result. In addition to the pig weight, the labors measure the body length and girth of the pig to know how much changes in pig growth. It is difficult for labors to control the pigs during the measurement. At least two labors are needed to control the pig body. In case of manual measurement, the pig body must be straight to get the stable result since the pig takes different poses. Thus, the mouth of the pig is fixed with the steel wire to avoid pig movement during the pig measurement. That is a hard work for both labors and pigs in every pig farms. In order to cope with these problems, we have developed the handy type measurement system to get parameters to estimate the pig weight by just capturing the image of the pig in the pig farm. The pig body length is defined as the length between the head and tail along the spine of the pig body. The position behind the fore legs of the pigs is known as girth position of the pig. The size of a pig body area is also important in pig weight estimation. These parameters are extracted automatically by our system, regardless of the posture of the pig.
In our system, Asus Xtion-2 Device is used to estimate pig weight. Laser slit is also used to align the direction of the pig body. Xtion-2 device contains RGB-D sensor and can provide 5M RGB resolution. Thus, clear depth image is captured with that device. That captured image is used as data in the estimation of pig weight.
Therefore, that system not only reduces works and time for labors in the pig farm but also releases pig struggling when guiding the pig to the load cell. After capturing depth image using Xtion-2 device, pig body to be measured is extracted automatically from that captured image. In extraction process of the pig body, Region Growing method is applied in our system. Region growing method can extract the target body robustly from the depth image. After extraction of the pig body, our system detects both 2-D data and 3-D data such as body length and girth to estimate the pig weight.
After extracting the parameters of the pig body of the captured image, Random Forest Algorithm, one of the machine learning method is applied to estimate the pig weight in our system. There are advantages of Random Forest Algorithm. Random Forest can be used for identifying the most important features from the training dataset. Therefore, Random Forest Algorithm is the appropriate method for measuring the pig weight in the practical condition of a pig farm. The estimated pig weight is accurate with the ground truth weight with the use of random forest method. The operator can know the estimated pig weight immediately by just capturing one image for the pig to be measured.
In our system, Asus Xtion-2 Device is used to estimate pig weight. Laser slit is also used to align the direction of the pig body. Xtion-2 device contains RGB-D sensor and can provide 5M RGB resolution. Thus, clear depth image is captured with that device. That captured image is used as data in the estimation of pig weight.
Therefore, that system not only reduces works and time for labors in the pig farm but also releases pig struggling when guiding the pig to the load cell. After capturing depth image using Xtion-2 device, pig body to be measured is extracted automatically from that captured image. In extraction process of the pig body, Region Growing method is applied in our system. Region growing method can extract the target body robustly from the depth image. After extraction of the pig body, our system detects both 2-D data and 3-D data such as body length and girth to estimate the pig weight.
After extracting the parameters of the pig body of the captured image, Random Forest Algorithm, one of the machine learning method is applied to estimate the pig weight in our system. There are advantages of Random Forest Algorithm. Random Forest can be used for identifying the most important features from the training dataset. Therefore, Random Forest Algorithm is the appropriate method for measuring the pig weight in the practical condition of a pig farm. The estimated pig weight is accurate with the ground truth weight with the use of random forest method. The operator can know the estimated pig weight immediately by just capturing one image for the pig to be measured.