Research And Application Of Dynamic Intelligent Perception System For Livestock In Smart Farm

Under the background of improving the quality of livestock products and increasing environmental pressure,livestock husbandry is gradually returning to the green ecological culture model.Smart farm is the new development trend of livestock husbandry,so our government strongly supports the development of intelligent equipment for livestock husbandry.At present,wild livestock information collection faces the problems of limited perceptual ability,difficult data collection,high power consumption and short power supply time,which are important obstacles to the overall trend of intelligent farm management.So,it is of great significance to study the information perception methods of livestock's physiological signs,convenient data interaction methods,and long-term information monitoring to improve decision-making and intelligent management levels.This paper addresses the need for long-term and convenient information collection under the free-range model in field,integrates technologies such as LoRa,energy harvesting and 3D printing,and develops a data exchange platform suitable for restocking mode with low power consumption,self-powered,long-distance and simple network structures.Cattle is the main livestock product for meat consumption in China.This paper focuses on the important clinical features of bovine physiology diseases--eating and ruminating.The chewing which is the common feature for eating and ruminating,combined with signal processing and artificial neural network technology are used to develop the chewing perception,frequency extraction and classification methods for cattle.The main research results of this article are reflected in the following aspects:(1)Based on the research of the process of chewing behavior of cattle and the principle of non-contact angle measurement,the movement of the upper jaw and the lower jaw during the chewing process of cattle is converted into angular movement for measurement.The non-contact magnetic encoder AS5600 is selected as the angle sensing chip,which isolates the mechanical part from the electronic part,indirectly measures the chewing process of cattle,and designs a chewing measurement-aided wearable accessory in combination with 3D printing technology.The experimental results show that the correct rate of chew measurement can reach to 98.4%.(2)On the basis of studying the characteristics of chewing behavior of cattle,the chewing data and head posture data are selected as effective chewing classification source data.After sampling processing,the chewing data is smooth-smooth filtered to get the peak and the trough was used to obtain homodyne of the peak and the trough.The chewing frequency is obtained by FFT.The head posture data is averaged.By putting the BP neural network in MATLAB software to learn,when it is near to the expected results,the weights and thresholds of the network are introduced into the microcontroller,and the BP neural network classification algorithm is implemented therein.The experimental results show that the correct rate of non-chewing can reach to 100%,the correct rate of eating is 70.7%,the correct rate of rumination is 92.3%,and the correct frequency of is 93.96% which could be provided as the quantification value of behavior.(3)The SMT32 cubmx development platform is used in this paper.A low-power microcontroller STM32L152 RC is used as the main control chip,and LoRa's SX1278 chip is used to realize the star network networking development to build the low-power,long-distance data collection mode;and solar energy is used as an energy source and is integrated with the BQ25504 energy harvesting chip,which expands the conversion efficiency of solar energy,so it reduces the device volume,easy to wear,and enables long-term power supply for the equipment.The innovations of this paper are as followed:(1)The chewing sensor for cattle developed in this article belongs to the original innovation,which simplifies the complex chewing process to angle measurement and simplifies the perception structure,with high accuracy and strong anti-interference characteristics.It provides an effective chewing measurement method for scientific research and production applications.(2)In this paper,BP neural network is used as the core algorithm of chewing classification.Based on limited microcontroller resources,online real-time processing of BP neural network is realized which improves the intelligence level of the node and reduces the power consumption of the device.When eating and rumination are classified,chewing frequency can also be provided as quantitative data of behavior to improve decision-making ability.(3)In this paper,a wireless data transmission system with the advantages of simple structure and wide coverage is developed.The remote monitoring of different animals and different physical signs can be realized by changing the sensors and wearing modes.