Research On Key Technologies Of Intelligent Perception Of Breeding Goose Body Mass

At present,the weighing of breeding geese in China is still dominated by traditional manual weighing.Frequent artificial weighing behavior in the breeding process will cause problems such as reduction in food intake,disorder of egg laying cycle,cross infection of human and poultry,large consumption of manpower and time and so on.According to the practical needs of intelligent non stress body mass perception of breeding geese,this paper develops and studies the key technologies of intelligent body mass perception of breeding geese,which provides technical support for large-scale,intelligent and efficient breeding of breeding geese.The main research contents of this paper are as follows:（1）The high-speed industrial camera was used to collect the walking video data of breeding geese,and an improved optical flow method based on weighted guided filtering（WGF）was proposed as the segmentation method of breeding goose contour.The morphological analysis of the collected goose motion video was carried out to divide the different walking processes of breeding geese（standing period,double support period,walking period,etc.）,so as to provide a theoretical basis for the analysis of gait parameters such as gait frequency and stride length of breeding geese.（2）According to the law of pressure distribution at the bottom of palm of goose,the pressure model at the bottom of palm based on movement is constructed,and a weighing method based on array piezoresistive thin film sensor is designed.The force analysis is carried out by using ANSYS software,and the scheme of array piezoresistive thin film sensor is optimized.By analyzing the influencing factors such as line arrangement angle and distance,the optimal line distribution characteristics are given,and the effectiveness and accuracy of the array piezoresistive thin film sensor are verified by experiments.（3）Aiming at the problem that the gait characteristics of breeding geese are difficult to be expressed in data,a gait feature extraction method based on the gait and joint distribution characteristics of breeding geese was proposed.The complete gait trajectory of breeding geese was quantitatively studied.According to the offset angle law of goose joint spacing and the distribution characteristics of contour pixels in continuous images,an improved dense trajectories（IDT）recognition algorithm was proposed.Experiments verified the reliability of the recognition algorithm which integrates the dynamic and static behavior information of geese.（4）Aiming at the problem that it is difficult to distinguish breeding geese in the natural environment of goose house,a gait classification algorithm based on improved spatial pyramid pooling（ISPP）convolution neural network was proposed.The adaptive separation adhesion method was used to obtain the complete and reliable breeding goose contour input,and the pyramid classification layer was used to replace the full connection layer in the convolution neural network to realize the semantic fusion between different feature layers in the breeding goose image,It can accurately obtained the individual identity information of breeding geese in high-speed frame in motion video,and the feasibility of the algorithm was verified by experiments.（5）By analyzing the influence of characteristic parameters such as the swing speed of each joint and the input of pressure signal on the weight perception accuracy of breeding geese in the process of breeding geese movement,a weight calculation method based on breeding geese movement model is proposed,the intelligent perception system of breeding geese body mass is developed,and the intelligent perception experiment of breeding geese body mass is carried out.The experimental results show that the identity information of breeding geese can be accurately identified in the perspective,and the average error rate of breeding geese body mass is 6.58%,The accuracy of body mass perception is high.