| The behavior information of breeding cocks is the external manifestation of their physical health,and the perception and analysis of individual behavior of breeding cocks is an inevitable requirement for accurate breeding.At present,this research mainly relies on manual observation,which is time-consuming,laborious and low efficiency.By identifying individual behavior through automation,it can improve the accuracy and efficiency of the research on cock behavior,and it is very significant to promote the development of Chinese livestock and poultry industry.In this paper,based on the research object of this cage breed rooster,the nine-axis acceleration sensor technology was adopted,combined with the optimization method of behavior characteristics,to establish the automatic behavior recognition model of breed rooster,which can accurately identify seven behaviors including fighting,mating,feeding,feathering,drinking,wing beating and pecking,providing technical support for the behavior monitoring and management of breed rooster.The main research contents of this paper include:(1)The establishment of the data set of behavioral characteristics of cage bred cocks.In this paper,the behavior data collection experiment was designed,and the best position of sensor wearing was determined through the preliminary experiment.The sliding average filter was used to reduce the noise of the data,and then the sliding window was used to extract the 44 dimensional features in the time domain and frequency domain contained in the acceleration and angular velocity.Finally,the data were normalized to eliminate the dimensional effects,and the behavioral feature data set of the roosters in the cage was established.(2)Construction and improvement of feature optimization model.By comparing different machine learning classification algorithms and dimensionality reduction methods,Whale Optimization Algorithm(WOA)combined with eXtreme Gradient Boosting(XGBoost)were selected to build the optimization model of behavioral recognition features for roosters.An Improved Whale Optimization Algorithm(IWOA)was proposed to solve the problem that the accuracy of the original whale optimization algorithm was insufficient and the performance of the behavior recognition model of the roosters was affected.The convergence rate and global searching ability of whale optimization algorithm were enhanced by introducing the best point set,adaptive weight and the learning strategy of dimensional-per-dimension lens imaging based on adaptive weight factor,and the optimization model of roosters behavior recognition features was established based on IWOA-XGBoost.(3)Results and analysis of individual behavior recognition of cocks.The IWOAXGBoost model was used to identify and verify the seven behavior data sets of the breed rooster.The results showed that the optimization ability and speed of the improved whale optimization algorithm were enhanced,and the 44-dimensional features were reduced to 11dimensional features after the feature optimization,which greatly reduced the model calculation.The accuracy rates of eating,drinking,fighting,wing beating,mating,feathering and pecking were 95.22%,90.68%,96.87%87.72%,68.39%95.79%and 88.81%,respectively.The accuracy rate of the model was 93.87%.The feature optimization results can improve the recognition accuracy of different classifiers and have certain universality.The research on the relationship between behavior recognition results and group order showed that pecking behavior decreased significantly with the decrease of group order(P<0.01),fighting behavior decreased significantly with the decrease of group order(P<0.1),and other behaviors had no significant relationship with group order(P>0.1).To sum up,the whale optimization algorithm model proposed in this paper based on the improved hybrid strategy can screen out effective classification and recognition features and has a high behavior recognition accuracy,which provides a reference for welfare precision farming. |
PDF Full Text Request