Research On Health Status Monitoring Technology Of Laying Hens In Facilities

Layer breeding not only meets the basic health needs of human beings,but also is the economic source of people’s production and living.It has a close relationship with people’s living standard.As the laying mother,the healthy constitution of laying hens not only affects the benefit of laying eggs,but also concerns the food safety of human beings.It is particularly important to carry out timely and effective health status detection of laying hens.In recent years,the rapid development of deep learning has brought innovation to many fields.Artificial intelligence technology has become more and more perfect.Aiming at the problems of poor robustness and low real-time performance of the existing laying hens health detection technology and death detection algorithm,this paper proposed to use computer vision technology to monitor the health status of laying hens.The thesis mainly focuses on the following aspects:Aiming at the problem of difficult identification of sick chickens caused by large number of chickens in large farms,a deep learning based method for digestive tract health status identification of laying hens was proposed.According to the different physiological state of digestive tract of sick and healthy chickens,the method of identifying chicken feces was used to judge the health state of digestive tract of chickens.In order to solve the problem of high missing rate in feces detection based on Faster R-CNN algorithm,an improved network structure is proposed.Firstly,based on the idea of feature pyramid,a multi-resolution feature fusion layer is introduced into the pooling layer of regions of interest,combining low-level location information with high-level semantic information to extract more convolution features of candidate regions,which solves the problem that the model needs to detect target information of different scales,and improves the phenomenon of missing detection.Secondly,the characteristics of the data were analyzed,and the optimized regional proposal network was proposed.A small anchor frame was used to match the target,and Mish function with smoothing characteristics was used as the activation function of the feature extraction network to avoid gradient saturation when the training speed dropped sharply.The experiment showed that the detection accuracy of the improved model F-Faster R-CNN reached 97.08%,the recall rate reached 93.14%,and the detection time was 0.2 seconds,which could accurately identify the feces of sick chickens and thus distinguish the health condition of laying hens.Aiming at the problem that dead chickens are easily disturbed by live chickens and difficult to detect,a dead chicken detection algorithm based on semantic segmentation method was proposed.According to the characteristics of the dead chickens,the semantic segmentation algorithm was used to locate the chicken body,and the chicken images were sampled twice at an interval of 10 minutes.The intersection ratio of the segmentation results of the two images was calculated and compared with the threshold value to identify the dead chickens.In order to solve the problem of unclear edge segmentation based on Seg Net algorithm,an improved model structure was proposed.Firstly,the hollow space pyramid pool is used to capture multi-field context information in the network structure.Secondly,the multi-layer depth separable convolution is used to replace the standard convolution layer,which reduces the computation and improves the real-time performance,enhances the ability of the network to extract depth features,and further improves the segmentation effect.Finally,a three-scale attention cascade fusion module is proposed to increase the weight of the target region and embed it between the Seg Net encoder and decoder in parallel to enrich the detailed features of the decoder and improve the accuracy of pixel segmentation.After correcting the second image of the chicken coop according to scale invariant feature transformation and K-nearest neighbor algorithms,the improved model AT-Seg Net was used to segment it.The intersection ratio of the segmentation results was calculated.When the ratio was greater than0.4,the chicken was determined to be dead.Finally,the accuracy rate of dead chicken detection was 96.5%.The error rate was 1.08%,which achieved a good detection effect.According to the actual management needs of breeding farms and combining with the detection algorithm proposed in this paper,a health status monitoring system of laying hens was developed,which could provide real-time interface of monitoring and detection.Firstly,the functional requirements of the system are analyzed.The visual interface of the monitoring system is developed with Py Qt tool.Functional modules are added.Secondly,the two detection algorithms are embedded in the visual interface,and different algorithm detection threads are called through module selection.Finally,the health status of chickens is displayed through the indicator light,and the background detection results are recorded to the local.The developed monitoring system helps the staff realize the online management of the farm.