Accurate flower-load assessment is critical for informed thinning strategies in orchard management. UAV-based deep learning automated counting offers efficiency advantages, yet precise counting is heavily dependent on abundant annotated data, which is scarce and costly to obtain in agricultural settings. While semi-supervised learning alleviates dependency on manual annotation, its application to UAV-based orchard imagery faces challenges: complex backgrounds and small target sizes, which undermine pseudo-label reliability. To address these challenges, this study proposes a two-stage framework to achieve separate counting of apple flowers at different phenological stages. First, a color-SAM flower extractor (CSAM-FE) is proposed to preprocess images using a strategy combining color thresholding with the Segment Anything Model (SAM), suppressing background noise and extracting high-quality flower clusters, thereby providing purified inputs for the subsequent counting network. Second, an uncertainty-guided semi-supervised flower counting network (USCount-Net) is proposed for accurate stage-specific flower counting with limited labeled data. The USCount-Net incorporates two key components: an adaptive pseudo-label filtering (PLF) mechanism based on frequent forward uncertainty estimation (FFUE) is designed to dynamically suppress noisy gradient backpropagation, mitigating error propagation from unreliable pseudo-labels; and a noise-sensitive adaptive gated fusion (AGF) module is introduced to fuse cross-scale features without redundancy, addressing significant scale variations across phenological stages and observation angles. Comparative experiments on a self-built apple flower counting dataset demonstrate that USCount-Net achieves lower MAE and RMSE than state-of-the-art methods at 10%, 30%, and 50% labeling ratios. The results demonstrate that the proposed methodology serves as methodological support for rapid and precise apple flower counting in low-annotation agricultural scenarios.

The new shoot density of slash pine serves as a vital indicator for assessing its growth and photosynthetic capacity, while the number of new shoots offers an intuitive reflection of this density. With deep learning methods becoming increasingly popular, automated counting of new shoots has greatly improved in recent years but is still limited by tedious and expensive data collection and labeling. To resolve these issues, this paper proposes a semi-supervised counting network (MTSC-Net) for estimating the number of slash pine new shoots. First, based on the mean-teacher framework, we introduce the improved VGG19 to extract multiscale new shoot features. Second, to connect local new shoot feature information with global channel features, attention feature fusion module is introduced to achieve effective feature fusion. Finally, the new shoot density map and density probability distribution are processed in a fine-grained manner through multiscale dilated convolution of the regression head and classification head. In addition, a masked image modeling strategy is introduced to encourage the contextual understanding of global new shoot features and improve the counting performance. The experimental results show that MTSC-Net outperforms other semi-supervised counting models with labeled percentages ranging from 5% to 50%. When the labeled percentage is 5%, the mean absolute error and root mean square error are 17.71 and 25.49, respectively. These findings demonstrate that our work can be used as an efficient semi-supervised counting method to provide automated support for tree breeding and genetic utilization.