We found a match
Your institution may have rights to this item. Sign in to continue.
- Title
基于 Anchor-free 网络结构的草菇生长状态检测.
- Authors
查磊; 仝宗军; 邵帅; 李正鹏; 余昌霞; 杨焕玲; 郭倩
- Abstract
A Resnet-50+FPN+RETINANET network was used to monitor the growth status of Volvariella volvacea during cultivation. Photos of V. volvacea were taken at different factories with different cultivation modes and different growth stages. Abnormal fruiting bodies and bacterial contamination were also taken photos. A total of 895 original photos were manually labelled using labellmg and classified to seven groups as follows: growing stage, maturation stage, pileus opening stage, miscellaneous bacteria contaminated V. volvacea, navel-shaped V. volvacea, Coprinopsis sp., and miscellaneous bacteria contaminated cultivation substrate. The model was first trained by the 895 photos to recognize different growth status, and then optimized by incorporation of modules including online hard example mining module OHEM, deformable convolution, model loss function improvement module focal-loss, non-local module for receptive field expansion and data augmentation module for training sample enrichment. The optimized model was finally tested under different illumination conditions. The results showed that the mean average precision(mAP) of the optimized model reached 83.7. The model can output number, current coordinates, type, RGB mean value, relative height and width, and can identify V. volvacea at different stages, fruiting body contamination, abnormal fruiting bodies, C. atramentaria, and cultivation substrate contamination.
- Subjects
BACTERIAL contamination; FRUITING bodies (Fungi); DATA augmentation; COMPUTER vision; DEEP learning
- Publication
Acta Edulis Fungi, 2022, Vol 29, Issue 2, p31
- ISSN
1005-9873
- Publication type
Article
- DOI
10.16488/j.cnki.1005-9873.2022.02.004