Collection of chicken behavior big data (mass data) has begun.The mainstream chicken is made of cage, and it is difficult to distinguish the inside with a camera because the chicken lines on the front of the cage when eating food.On the other hand, the mainstream of meat broilers is flat.If you are flat, you can collect more than 100 chicken behavioral data with one camera installed on the ceiling.
Professor Mitsuhiko Katahira and Professor Kenichi Horiguchi of Yamagata University, and Shigeru Ichiura of VIAR & E (Minato -ku, Tokyo) focus on the development of chicken behavior judgment technology using artificial intelligence (AI) technology.
By learning 4000 images, the identification rate of breeding, drinking water, and rest was 100 %.
"I am trying a method that other researchers do not choose. If a central operational treatment device (CPU) specializes in AI comes out, learning will not be difficult."Ichiura is from Nvidia, a major US semiconductor.We are collecting data in anticipation of future CPUs.
At the same time, the technology of tracking individuals and tracking each bird with a wing crest is under development.Record the speed of gaining weight for each bird and visualize the amount of activity.Chicken has a habit of creating a harsh order, such as poking another chicken and peeling the blades.Therefore, if you can identify a solid and weak solid in the group, you can respond to the parcel.Professor Horiguchi hopes that the observation eyes of poultry farmers can be used in AI.
Associate Professor Makoto Ozawa of Kagoshima University has developed a technology to estimate the weight and amount of chicken from camera images in collaboration with Fujitsu Kagoshima Infonnet (Kagoshima City) and Nakajima Seisakusho (Nagano City).Image recognition AI identifies chicken, finds a three -dimensional volume from the size of the chicken in the image, and calculates weight.The estimated error boasts a 5 % plus or minus accuracy before adjustment.
In the poultry house, the weight gains at a constant pace from putting chicks to shipment.Associate Professor Ozawa explains, "There is an ideal weighing speed for each farm. We determine how close the ideal value is and determine the shipment date."
In the measurement of the amount of activity, the chicken is identified and calculated how much it moved in a few seconds later.Find a sign that chicken will die due to illness, such as catching a cold and lowering the amount of activity.At present, the cause has been distinguished after the dying chicken has increased.Generally, broilers fall as they get fat.In response to daily changes, it is also a sign that something abnormal is occurring when the amount of activity suddenly decreases.
Currently, Nakajima is developing an image recognition system for practical use.The company plans to work with the company's data collection system and use it for daily breeding management.
Nikkan Kogyo Shimbun July 15, 2021
Comment
Cameras, AI, and flat keeping are compatible and large amounts of data can be collected.First of all, the introduction should be advanced as a physical condition detection where the needs have become apparent.Next is analysis of action needs and improvement of breeding environment.I think there is a breeding history guarantee that considers animal welfare.If this is recognized as a brand or added value, business will be established.Unless you use animals and creatures, human society will not hold.Even if the rights of animals are difficult to fulfill, I think that animal welfare can be a business rational activity.