Broiler weighing is essential in the broiler farming industry. Camera-based systems can economically weigh various broiler types without expensive platforms. However, existing computer vision methods for weight estimation are less mature, as they focus on young broilers. In effect, the estimation error increases with the age of the broiler. To tackle this, this paper presents a novel framework. First, it employs Mask R-CNN for instance segmentation of depth images captured by 3D cameras. Next, once the images of either a single broiler or multiple broilers are segmented, the extended artificial features and the learned features extracted by Customized Resnet50 (C-Resnet50) are fused by a feature fusion module. Finally, the fused features are adopted to estimate the body weight of each broiler employing gradient boosting decision tree (GBDT). By integrating diverse features with GBTD, the proposed framework can effectively obtain the broiler instance among many depth images of multiple broilers in the visual field despite the complex background. Experimental results show that this framework significantly boosts accuracy and robustness. With an MAE of 0.093 kg and an R² of 0.707 in a test set of 240 63-day-old bantam chicken images, it outperforms other methods.

In this study, we present a novel concept termed open-vocabulary domain generalization (OVDG), which we investigate within the context of semantic segmentation. OVDG presents greater difficulty compared to conventional domain generalization, yet it offers greater practicality. It jointly considers (1) recognizing both base and novel classes and (2) generalizing to unseen domains. In OVDG, only the labels of base classes and the images from source domains are available to learn a robust model. Then, the model could be generalized to images from novel classes and target domains directly. In this paper, we propose a dual-branch FreeMix module to implement the OVDG task effectively in a universal framework: the base segmentation branch (BSB) and the entity segmentation branch (ESB). First, the entity mask is introduced as a novel concept for segmentation generalization. Additionally, semantic logits are learned for both the base mask and the entity mask, enhancing the diversity and completeness of masks for both base classes and novel classes. Second, the FreeMix utilizes pretrained self-supervised learning on large-scale remote-sensing data (RS_SSL) to extract domain-agnostic visual features for decoding masks and semantic logits. Third, a training tactic called dataset-aware sampling (DAS) is introduced for multi-source domain learning, aimed at improving the overall performance. In summary, RS_SSL, ESB, and DAS can significantly improve the generalization ability of the model on both a class level and a domain level. Experiments demonstrate that our method produces state-of-the-art results on several remote-sensing semantic-segmentation datasets, including Potsdam, GID5, DeepGlobe, and URUR, for OVDG.

In this study, we present a novel concept termed open vocabulary domain generalization (OVDG), which we investigate within the context of semantic segmentation. OVDG presents greater difficulty compared to conventional domain generalization, yet it offers greater practicality. It jointly considers 1) recognizing both base and novel classes and 2) generalizing to unseen domains. In OVDG, only the labels of base classes and the images from source domains are available to learn a robust model. Then the model could be generalized to images from novel classes and target domains directly. In this paper, we propose a dual-branch FreeMix module to implement the OVDG task effectively in a universal framework: the Base Segmentation Branch (BSB) and the Entity Segmentation Branch (ESB). First, the entity mask is proposed for the first time for segmentation generalization, and the semantic logits are learned for both base mask and entity mask, so that to enhance the diversity and completeness of masks for both base classes and novel classes. Second, the FreeMix utilizes pre-trained self-supervised learning on large-scale remote sensing data (RS_SSL) to extract domain-agnostic visual features for decoding masks and semantic logits. Third, a training tactic called dataset-aware sampling (DAS) is introduced for multi-source domain learning, aimed at improving the overall performance. In summary, RS_SSL, ESB and DAS can significantly improve the generalization ability of model on both class-level and domain-level. Experiments demonstrate that our method produces state-of-the-art results on several remote sensing semantic segmentation datasets, including Potsdam, GID5, DeepGlobe, and URUR, for OVDG.