Research On Sound And Posture Recognition Of Live Pigs Based On Intelligent Algorithms

The convergence of artificial intelligence and intelligent monitoring technology is becoming a new application path as the domestic porcine farming automation and intelligence levels continue to grow.More and more researchers have started relevant research.Currently,regular disease control of pigs is mostly accomplished by manual observation and screening of irregular pigs,which is both time-consuming and labor-intensive.In addition,the accuracy is poor due to the subjective factors of the keepers.At the same time,the activities of the personnel will cause the stress response of the pigs,which is not conducive to the healthy growth of the pigs.Therefore,the integration of artificial intelligence algorithms and intelligent monitoring technology to replace traditional manual observation methods will be a major trend in the developmen t of domestic porcine farming.In this background,this thesis used intelligent algorithms to initially realize the effective classification and recognition of the typical sounds and postures of pigs.Finally,a set of monitoring software based on MATLAB＿G UI was designed,and the state of pigs is monitored by calling sound and image re cognition algorithms.The work done in this thesis is as follows:(1)The sound and images of pigs are collected through directional pickups and gun-type cameras.Due to the complex environment of the Pigsty and there is a lot of noise,this thesis preprocesses the sound and image separately,uses Wiener filtering and improved wavelet threshold denoising metho d to denoise the sound signal,and the samples with better noise reduction effects are selected to continue preprocessing such as endpoint detection and frame windowing;the image is enhanced by bilateral filtering,and the outline of the target pig is segmented using the Otsu method,and finally a binary image of the pig is obtained based on morphological processing.(2)To judge the abnormal state from the sound of live pigs,a recognition algorithm for the abnormal sound of live pigs was studied.The method firstly extracts the characteristic parameters of live pig sound signal,including short-term energy,short-term zero-crossing rate,and Mel frequency cepstrum coefficient.Then,based on the characteristics of the pig sound signal,different feature combinations were selected to establish a one-dimensional convolutional neural network model and a DE-VRF model.Through experimental comparison and analysis,after the feature is reduced by PCA,the DE-VRF model has the best recognition effect,reaching 96.76%.(3)To better recognize the posture of live pigs,a pig postur e recognition algorithm based on the firework algorithm to optimize ELM is studied.The method first extracts the 7-dimensional Hu invariant moments of the pig and the 4-dimensional gray-level co-occurrence matrix for feature fusion.Then use the firework algorithm to optimize the hidden layer bias and weight of the extreme learning machine,and finally input the11-dimensional feature quantity into the FWA-ELM model for recognition.Through experimental comparative analysis,the recognition rate of FWA-ELM reaches 96.89%,which is higher than that of DE-ELM,PSO-ELM,and other models.(4)MATLAB＿GUI is used as the tool platform to develop software with functions of pig sound recognition and pig posture recognition.The function of each module is realized by calling the algorithm,and the research algorithm is finally tested.