Live pig counting method based on instance segmentation

Abstract

The invention belongs to the field of computer vision and machine learning, and discloses a live pig counting method based on instance segmentation. The defects that an existing algorithm is high in labeling cost, low in precision and poor in robustness are overcome. According to the technology, firstly, a category prediction result of each pixel in an image is obtained by utilizing an image segmentation network, then, image segmentation loss, instance splitting loss, positioning loss and regression loss are respectively calculated, and finally, each loss is subjected to weighted fusion and is used as supervision information to train the network. According to the technology, only the central points of the pigs need to be marked, an approximate segmentation area can be output for each pig instance, then the accurate position and number information of the pig instances is obtained, and the pig counting and positioning problems in a complex environment are solved.

Description

technical field [0001] The invention belongs to the field of computer vision and machine learning, and in particular relates to a method for counting live pigs based on instance segmentation. Background technique [0002] Pig counting is a critical task in large-scale farming management. Accurate pig counts can not only improve the management level of pig breeding and pig house construction, but also help reduce breeding costs, detect abnormal situations such as missing pigs in a timely manner, and improve the competitiveness of farms. However, currently counting pigs still relies on manual counting, which not only requires a lot of labor costs, but also is inefficient. For example, it usually takes about three days to count pigs in a medium-sized farm. In addition, the number of pigs changes frequently due to factors such as frequent trading and sow production, making the task of pig counting more difficult. [0003] In recent years, various computer vision algorithms hav...

AI Technical Summary

Problems solved by technology

However, the current counting of pigs still relies on manual counting, which not only consumes a lot of labor costs, but also is inefficient. For example, it usually takes about three days to count the number of pigs in a medium-sized farm

Furthermore, due to factors such as frequent trading and sow production, the number of pigs changes frequently, making the task of counting pigs even more difficult

[0003] In recent years, various computer vision algorithms have been widely used in the field of agriculture and agricultural automation, but there are still two problems in practical application: 1. Due to the overlapping occlusion of pig groups, light fluctuations, similar appearance of pigs, shape deformation and obstacles Object occlusion and other issues limit the use of traditional computer vision techniques in pig counting tasks

2. In order to achieve the goal of counting objects, the existing methods based on deep learning usually need to rely on object detection or image segmentation technology, which require bounding box labeling and pixel-level labeling of objects respectively. Responding to scenarios with highly dense pig herds and small pig targets

Images