A lactating sow posture conversion identification method based on Faster R-CNN and HMM

Abstract

The invention discloses an R-based on a Faster. The CNN and HMM lactating sow posture conversion identification method comprises the following steps: 1, enhancing the quality of a depth image; 2,usingan improved Faster R-CNN to identify the postures of the sows, and take the posture with the maximum probability per frame as a posture sequence; taking the first five detection frames with the maximum probability as candidate areas; 3, correcting an attitude sequence classification error by using median filtering with the length of 5; detecting a suspected conversion segment by using the video segment attitude conversion times; In the suspected conversion segment, constructing a sow positioning pipeline according to the candidate area by using a Viterbi algorithm; 4, in the positioning pipeline, segmenting each frame of sow by using a maximum between-class variance method, and calculating the height of each part of the sow body to form a height sequence; 5, inputting the height sequenceinto an HMM model, and dividing a suspected conversion segment into an attitude conversion segment and an unconverted segment; and classifying the single posture segment and the posture conversion segment to obtain a recognition result. The sow posture recognition system can automatically convert and recognize the posture of the sow under the conditions of light change and nighttime, and lays a foundation for high-risk behavior recognition.

Description

Technical field [0001] The present invention relates to the technical field of video recognition, and in particular to a posture conversion recognition method for suckling sows based on Faster R-CNN and HMM. Background technique [0002] In the concentrated feeding environment of the pig farm, the maternal behavior of sows is closely related to the survival rate of piglets, and the quality of maternal behavior is mainly reflected in the transformation of posture. Observing the posture transition behavior of sows through artificial eyes or video surveillance is highly subjective and time-consuming and labor-intensive. The automatic recognition of sow posture conversion behavior can provide basic research information on the characteristics and laws of its maternal behavior, prevent piglet death from trampling, improve piglet survival rate, reduce labor cost of pig farm management, and have great significance for improving the level of pig breeding. [0003] Sensor technology has bee...

AI Technical Summary

Problems solved by technology

However, body deformation during sow posture transition, adhesion between sow and piglet, and scene lighting changes have brought great challenges to all-weather computer vision-based sow posture transition recognition. Therefore, there are almost no reports of relevant research results at home and abroad.

Images