Research On Pig Face Recognition Based On Cloud Edge Collaborative Computing

Due to the problems such as high cost,gradually blurred labeling and easy to cause wound infection,the existing pig labeling methods,such as ear tags and two-dimensional codes,are difficult to meet the needs of the development of intelligent livestock breeding.Therefore,it’s becoming a developing trend to apply recognition technology based on deep learning to livestock identification.In order to give consideration to the accuracy and speed of the pig face recognition task,this paper based on the cloud edge collaboration architecture and introduce some effective units into the YOLOv3 and YOLOv3-tiny models to design detection models applied to the cloud and edge respectively.The main contents are as follows:(1)Firstly,the implementation of cloud edge collaborative architecture.The cloud is responsible for saving the collected samples,training the test model,and deploying the test model for testing.The edge end is responsible for collecting samples and deploying lightweight models to detect individual pigs nearby.When the edge end encounters an undetectable sample,it transmits the sample to the cloud through the Internet,and the cloud saves the detection result after detection,and transmits the result to the edge end;(2)Secondly,the design of the model on cloud.In the design of the cloud model,based on the YOLOv3 model,a Dense module is introduced into the feature extractor to improve the feature extraction capability of the model.Aiming at the problem that it is difficult to detect small and medium-sized targets and targets with facial shielding in group culture environment,an improved SPP unit was introduced to integrate multi-scale features to improve the characterization ability of the model.The experimental results show that the improved cloud detection model can achieve detection of small targets and the detection accuracy is significantly improved.(3)Finally,the design of the edge end model.In the design of the edge model,the YOLOv3-tiny model was used as the basic model to select the detector size for the collected pig data set.Then the basic unit of Shufflenetv2 model is introduced into the selected detection model to improve the feature extraction ability of the model while reducing parameters.Finally,choose the appropriate width for the model.Experimental results show that the improved edge model improves detection speed by4 frames/s,detection accuracy by 0.17%,and decreases memory consumption by14.4MB compared with the Yolov3-tiny model.