Rubber powdery mildew, caused by the fungal pathogen Oidium heveae Steinm., is a prevalent disease in rubber plantation regions worldwide. This disease significantly impacts the growth and yield of rubber trees, leading to substantial economic losses within the rubber industry. In recent years, due to climate change and adjustments in planting structures, both the geographical spread and severity of the disease have increased. Consequently, there is an urgent need to develop efficient remote sensing monitoring methods for early warning and effective management. To fully exploit disease information within hyperspectral data, this study first extracted spectral features using three methods: spectral mathematical transformations (MT), continuous wavelet transformation (CWT), and vegetation indices (VIs). Subsequently, correlation analysis (CA), least absolute shrinkage and selection operator (LASSO), and principal component analysis (PCA) were employed to select optimal features from each set, resulting in the construction of nine independent basic feature sets. To further enhance model performance, features selected by these three strategies (CA, LASSO, and PCA) were combined to form three fused feature sets. Finally, all basic and fused feature sets were input into a Random Forest (RF) model to evaluate the impact of different feature combinations on the accuracy of disease severity classification. The results revealed that, among the spectral data processing methods, CWT performed the best. Among the feature selection methods, PCA was the most effective. The feature fusion methods significantly improved model performance. Specifically, the fused feature set based on PCA selection (PCA_ALL) achieved the highest classification accuracy, with an overall accuracy (OA) of 98.89% and a Kappa coefficient of 0.98. This OA was 8.89% higher than that of CA_ALL and 4.42% higher than the best-performing basic feature set (PCA_CWT). This study establishes a remote sensing monitoring framework for classifying rubber leaf powdery mildew severity based on the fusion of multi-dimensional hyperspectral features. This framework not only lays a technical foundation for the transition of the natural rubber industry from experience-based control to intelligent decision-making but also provides crucial parameters for large-scale dynamic disease monitoring using UAV and satellite platforms.

Early-stage, accurate and high-throughput phenotyping‌ through leaf area estimation is ‌critical‌ for future rapeseed breeding, but faces ‌two key constraints‌: expensive data annotation and persistent challenge of leaf occlusion. To address these issues, we present a ‌data-efficient‌ deep learning framework using smartphone-captured top-down RGB images for rapeseed leaf area quantification. Our approach utilizes a two-stage strategy where a Vision Transformer (ViT) backbone is first pre-trained on a large, aggregated dataset of diverse, non-rapeseed public plant datasets using the DINOv2 self-supervised learning method. This pre-trained model is then fine-tuned on a custom rapeseed dataset using a novel Canopy-Mix data augmentation technique to handle fragmented views analogous to occlusion, and a hybrid loss function combining Smooth L1 and Log-Cosh for robust convergence. Through rigorous 5-fold cross-validation, our proposed model achieved strong predictive performance (Coefficient of Determination, R[Formula: see text]=0.805). Moreover, the predicted leaf area demonstrated a remarkably strong correlation with both fresh weight (r=0.900) and dry weight (r=0.885). The model significantly outperformed a range of baselines, including models trained from scratch, those pre-trained on ImageNet, and a heuristic method based on manually annotated bounding boxes. Ablation studies confirmed the essential contribution of each component, while qualitative analysis of attention maps demonstrated the model's ability to precisely localize the leaf canopy and ignore background distractors. This study demonstrates that domain-specific self-supervised pre-training offers a powerful solution to overcome data limitations in agricultural vision, providing a robust and scalable tool for non-destructive phenotyping that can potentially accelerate the rapeseed breeding cycle.

Wheat ranks third among cereal crops in terms of global production, and its demand is expected to increase as the human population grows. Plant breeding can increase crop production without burdening natural resources, and one way to accelerate genetic gain is through shortening breeding cycles with speed breeding (SB). Speed breeding protocols for winter wheat have been adapted by adding a vernalization phase to existing spring wheat protocols. Although a protocol for the vernalization phase was previously developed, it was not tested for genotypes grown in the Midwest US, which may have higher vernalization requirements. The transition from vegetative to reproductive stages in winter wheat depends mainly on photoperiod, vernalization temperature, and vernalization length, which determines the time needed to reach flowering. Optimizing vernalization under SB in a greenhouse setting is important for applications in breeding programs. Our objectives were to develop a speed breeding protocol for winter wheat that meets the vernalization requirements of all genotypes and to evaluate the interaction between vernalization temperature and sowing depth.

Accurate measurement of key phenotypic traits, including the horizontal and vertical diameters, the weights of both fruit and pit, is essential for the selection of elite litchi cultivars and the advancement of breeding research. Manual measurement, however, is laborious, inefficient, and subjective, highlighting the urgent need for automated and precise phenotyping tools. Unlike apples, mangoes, and grapes, litchi combines a spiny, highly variable pericarp (heterogeneous areoles/tubercles across cultivars) with diverse seed morphology (including irregular, wrinkled aborted seeds), thereby increasing the difficulty of semantic segmentation and biasing diameters and weight estimation. This study presents LitchiPhenoNet, a multimodal learning framework for litchi phenotypic analysis that employs a dual-branch architecture integrating RGB (color/texture) and depth (spatial/structural) information. Experiments were conducted on an RGB-D dataset comprising 1,198 image pairs (1280×720) across 10 cultivars, using a stratified train/test split of 958/240 pairs by cultivar. To address inherent semantic and scale inconsistencies between modalities, the framework incorporates the RD-Fusion module for precise cross-modal feature extraction, improving robustness under complex and variable pericarp surfaces. Comparative experiments show that LitchiPhenoNet consistently outperforms leading YOLO-based models, achieving millimeter-level diameter estimation with coefficients of determination approaching 0.98 and mean errors within 2 mm. For weight estimation, gram-level precision is attained across whole fruit, pit, and pulp, with coefficients of determination up to 0.98 and mean errors comparable to repeated manual measurements. By handling fine-scale surface relief and cross-cultivar variability, the framework is readily extensible to other textured fruits and scalable for high-throughput phenotyping in breeding programs. Collectively, these results demonstrate that LitchiPhenoNet provides an efficient, reliable, and accurate solution for quantifying litchi phenotypic traits, substantially advancing the objectivity and efficiency of phenotypic analysis and breeding selection.

Bread wheat (Triticum aestivum L.) is a vital global crop, supplying 20% of the protein in the human diet. Improving its productivity and resilience, particularly under water-limited conditions, is a major breeding priority. Genomic selection offers a promising approach to accelerate genetic gains by predicting complex traits using genome-wide marker data. This study evaluated the performance of various genomic selection (GS) models in predicting key agronomic traits under contrasting well-watered (WW) and water-stressed (WS) conditions, with the goal of enhancing drought adaptation in wheat breeding programs.

Seed quality analysis using X-rays is increasingly explored due to its non-invasive and rapid nature. Yet, the current absence of reliable and standardised imaging protocols has led to contradictory effects of X-ray exposure in previous studies. Our work systematically investigated the effect of low-energy X-rays (peak energy ≲25 keV) with limited doses (< 3 mGy) on a wide range of plant materials.

Plant diseases can cause heavy yield losses in arable crops resulting in major economic losses. Effective early disease recognition is paramount for modern large-scale farming. Since plants can be infected with multiple concurrent pathogens, it is important to be able to distinguish and identify each disease to ensure appropriate treatments can be applied. Hyperspectral imaging is a state-of-the art computer vision approach, which can improve plant disease classification, by capturing a wide range of wavelengths before symptoms become visible to the naked eye. Whilst a lot of work has been done applying the technique to identifying single infections, to our knowledge, it has not been used to analyse multiple concurrent infections which presents both practical and scientific challenges. In this study, we investigated three wheat pathogens (yellow rust, mildew and Septoria), cultivating co-occurring infections, resulting in a dataset of 1447 hyperspectral images of single and double infections on wheat leaves. We used this dataset to train four disease classification algorithms (based on four neural network architectures: Inception and EfficientNet with either a 2D or 3D convolutional layer input). The highest accuracy was achieved by EfficientNet with a 2D convolution input with 81% overall classification accuracy, including a 72% accuracy for detecting a combined infection of yellow rust and mildew. Moreover, we found that hyperspectral signatures of a pathogen depended on whether another pathogen was present, raising interesting questions about co-existence of several pathogens on one plant host. Our work demonstrates that the application of hyperspectral imaging and deep learning is promising for classification of multiple infections in wheat, even with a relatively small training dataset, and opens opportunities for further research in this area. However, the limited number of Septoria and yellow rust + Septoria samples highlights the need for larger, more balanced datasets in future studies to further validate and extend our findings under field conditions.

Efficient regeneration protocols are essential for large-scale propagation and genetic manipulation of recalcitrant medicinal species such as Cannabis sativa. Existing direct and indirect regeneration methods are highly genotype and explant-dependent, limiting broader applicability. Here, we report a five-stage (S-S) optimised protocol that is reproducible and achieves high-efficiency direct de novo regeneration using cotyledonary node explants from both hemp and medicinal cannabis genotypes. A 1% (v/v) H₂O₂-based sterilisation method significantly improved seed germination and reduced endophyte contamination. Among embryo-derived explants, the cotyledonary node attached to the cotyledon showed superior regeneration efficiency through two distinct pathways: axillary shoot initiation and de novo regeneration, the latter achieving ~ 70-90% efficiency in six hemp cultivars and three medicinal cannabis lines on TDZ and NAA containing shoot regeneration medium. Histological analysis confirmed true de novo shoot formation from peripheral cortical cells, independent of pre-existing meristems or callus. De novo shoots were initiated within 2 d of shoot regeneration medium treatment, indicating rapid cellular commitment to organogenesis, with optimal regeneration between 7 and 14 d. Prolonged exposure proved detrimental, causing excessive callusing and vitrification. Repeated subculturing during proliferation stage enabled scalable shoot multiplication, yielding an average of 7 shoots per responding explant (~ 11.4 shoots per seed), outperforming previously published cotyledon-based (~ 2-fold) and hypocotyl-based (~ 5-fold) methods under comparable conditions. Regenerated plantlets developed healthy roots (with IAA or IBA) and acclimatised readily, exhibiting normal vegetative and reproductive growth. The protocol's reproducibility across diverse cannabis genotypes and its applicability to other medicinal angiosperm species in this study highlights its value for both research and commercial applications.

Field trials play an essential role in developing genetically modified and genome-edited biotechnology plants, as they assess plant growth, yield, and potential unintended effects. Australian biotechnology field trials are regulated by federal protocols that mandate accurate forecasting of flowering times. Currently, this relies on labour-intensive and subjective visual field inspections of individual wheat plants at defined growth stages (Zadoks growth stages Z37, Z39, and Z41). To enable automatic forecasting, hyperspectral and red-green-blue (RGB) images were captured in the greenhouse, and hyperspectral reflectance data were acquired in a semi-natural environment. Support Vector Machine classification achieved F1 scores (0.832) for pre-anthesis growth stage classification through the combined use and systematic comparison of three spectral transformations, including Standard Normal Variate, Hyper-hue, or Principal Component Analysis, which together outperformed reliance on any single transformation. After feature selection, F1 scores (0.752) could be achieved with only five wavelengths. Furthermore, the SNV transformation demonstrated robust performance under limited training conditions, maintaining high classification accuracy and strong generalizability across varying data sizes. These findings highlight the effectiveness of transformation-enriched data and optimized feature selection for accurate growth stage classification, providing a low-cost approach to reduce manual inspection burdens and strengthen biosafety during biotechnology field trial practices.

Variety mixtures combining crop varieties with different root system properties have the potential to improve soil exploration through belowground niche complementarity, which can improve soil resource acquisition and crop productivity. However, there is a lack of appropriate methods to distinguish and quantify roots of different varieties, which limits our ability to elucidate belowground processes that underpin soil exploration and resource uptake by plants in variety mixtures.

Plant reproduction involves dynamic spatiotemporal changes that occur deep within maternal tissues. In ovules of Arabidopsis thaliana (A. thaliana), one of the two synergid cells degenerates at fertilization, while the fertilized egg cell (zygote) undergoes directional elongation followed by asymmetric division to initiate embryonic patterning. However, morphological analysis of these events has been hampered by the limitations of conventional cell wall staining, which fails to label cells lacking complete walls, and by the requirement for transgenic fluorescent reporters to visualize cell outlines. Here, we report that the membrane-specific fluorescent dye FM4-64 readily permeates ovules, allowing clear visualization of reproductive cell morphology both before and after fertilization. This staining method supports high-resolution time-lapse imaging and quantitative analysis of early embryogenesis in living tissues. Importantly, it is applicable not only to the angiosperm A. thaliana but also to the liverwort Marchantia polymorpha (M. polymorpha) and the fern Ceratopteris richardii (C. richardii), enabling the visualization of live reproductive cell structures within maternal tissues and revealing fertilization-associated morphological changes. This simple and robust method thus provides a valuable tool for spatiotemporal and quantitative analyses of reproductive processes across a broad range of plant species, without the need to generate transgenic lines.

Plant growth and development strongly depend on the uptake of soil minerals and their distribution within plants. Various electrophysiological techniques have been developed to study these ion transport processes and the role of ions in signal transduction pathways. An important non-invasive method is provided by Scanning Ion-Selective Electrodes (SISE), which are used to detect ion fluxes. These SISE-measurements depend on software that coordinates the continuous electrode movement between two positions, as well as data collection and analysis. We developed two LabView-based programs; the SISE-Monitor and SISE-Analyser that enable ion flux recordings and their analysis, respectively. These applications are freely available, both as windows-executable files that enable routine measurements, as well as the LabView source code that allows insights into the routines used for measurement and analysis. Moreover, the source code can be used to develop new functions, such as the combined measurement of extracellular ion fluxes with SISE and cellular ion concentrations with fluorescent dyes, or proteins.

Pyricularia oryzae is a major fungal pathogen responsible for significant yield losses in rice. In recent years, diverse pathotypes have emerged as threats to other economically important grasses, including ryegrass, oats, wheat and foxtail millet. Research on host-pathogen interactions involving this species requires reliable spore production for inoculation. However, as a hemibiotrophic pathogen, P. oryzae often sporulates poorly on artificial media and typically requires specialized two-stage protocols for consistent spore production. Although several such methods have been developed, all were optimized for rice-derived strains and have not been systematically evaluated across strains from other hosts. There is also a practical need for a simple setup that allows advance preparation and frozen storage of spore stocks. Therefore, we developed a new two-stage filter paper method and compared it with four published protocols across 23 strains from 13 grass hosts.

Black huckleberry (Vaccinium membranaceum) is a native fruit species of high nutritional, medicinal, ecological, and economic value. The black huckleberries, abundant in bioactive compounds, offer significant antioxidants and anti-inflammatory effects and play a key role in maintaining wildlife and forest ecosystems. Despite its importance, protoplast isolation and gene editing have not been reported in this species. These techniques are essential for functional genomics and crop improvement, but the recalcitrant nature of this species, complex genome, and variable ploidy present significant challenges for cellular and molecular manipulation. This study aimed to establish a reliable protocol for efficient mesophyll protoplast isolation and transient gene expression in V. membranaceum using in vitro-grown leaves.

In agricultural images acquired under natural conditions, pomegranate fruits are often partially occluded by leaves and branches, resulting in missing structural information that compromises the accuracy of yield estimation and automated harvesting. To overcome the challenges of recovering structural integrity in occluded agricultural imagery, we propose the Conditional Segmentation-guided Diffusion Network (CSD-Net). CSD-Net is a lightweight, unified framework, representing the first conditional diffusion model specifically designed for the joint tasks of pomegranate image completion and segmentation. CSD-Net aims to address the structural fidelity limitations of traditional completion methods. It utilizes a shared encoder, a segmentation branch, and an RGB diffusion branch. Crucially, the network leverages the segmentation mask as a key structural prior condition to guide the diffusion generation process. This innovative conditional guidance mechanism ensures high-fidelity reconstruction of fruit structures while maintaining spatial and textural consistency. Experimental results demonstrate that CSD-Net substantially outperforms conventional methods across metrics, achieving 30.37 dB in PSNR and 0.9490 in SSIM. Furthermore, its model size is only 117 MB, striking an effective balance between high completion quality and inference efficiency. This study offers a novel and highly effective solution for mitigating occlusion issues in agricultural visual perception tasks. Upon acceptance of this paper, the source code will be made publicly available at https://github.com/zdkd/PCSN .

Cotton diseases and pests pose significant threats to cotton production, necessitating accurate and efficient classification methods. Despite existing advanced methods, there is a research gap in utilizing both local feature extraction and global context capture for enhanced classification accuracy. Hence, this study developed and evaluated three advanced models for cotton disease and pest classification: a convolutional neural network (CNN)-based model, a Vision Transformer (ViT)-based model, and a hybrid CNN-ViT model. These models were trained on a dataset comprising eight classes of cotton diseases and pests, namely aphids, armyworm, bacterial blight, cotton boll rot, green cotton boll, healthy, powdery mildew, and target spot. The results demonstrated that the hybrid CNN-ViT model achieved the highest overall performance with an average test accuracy of 98.5%. The CNN model showed strong performance with an average accuracy of 97.9%. The ViT models, while having self-attention mechanisms to capture context and dependencies, exhibited improved performance with increased depth. The ViT model having four transformer layers outperformed the two-layer variant, achieving an average accuracy of 97.2% compared to 96.3%. The hybrid model effectively combined the strengths of CNN's local feature extraction and ViT's global feature capture, resulting in superior classification accuracy across most classes. Future research should focus on expanding the dataset to include more diverse diseases and pests and integrating the models with autonomous platforms for spraying the chemicals, thus facilitating real-world adoption and application in agricultural settings.

Stomata regulate gas and water exchange in plants and are crucial for plant productivity and survival, making their trait analysis essential for advancing plant biology research. While current machine learning methods enable automated stomatal trait extraction, existing approaches face significant limitations that require extensive manual labeling for training and additional human annotation when applied to new species. This study presents an automated system for extracting stomatal traits from Pisum sativum (pea) leaves that addresses these challenges through generative artificial intelligence. Our pipeline integrates imaging, detection, segmentation, and synthetic data generation processes. A nail polish impression technique was employed to prepare leaf microscopic images, followed by the application of deep learning networks to identify and segment stomata in these images. By including generative AI-produced synthetic data, our system achieves high segmentation accuracy across species, reducing manual relabeling requirements. This approach enables seamless cross-species model adaptation for many cases, alleviating the annotation bottleneck that often limits machine learning applications in plant biology. Our results demonstrate the pipeline's effectiveness for automated stomatal trait extraction and highlight generative AI's transformative potential in advancing stomatal detection methodologies, offering a scalable solution for broad-scale comparative stomatal analysis.

Field-scale assessment of chili leaf curl complex presents a significant diagnostic challenge, as both chili leaf curl virus (ChiLCV) and mite infestations produce visually overlapping symptoms difficult to distinguish by untrained personnel. This diagnostic confusion frequently leads to inappropriate application of either insecticides or acaricides, resulting in economic losses and environmental concerns. To address this issue, we propose SCA-MobiPlant, an improved MobileNetV3-Small model integrated with a novel multistage Squeeze-and-Excitation Coordinate Attention (SCA) fusion mechanism, designed for accurate differentiation of these apparently similar symptoms and precise field assessment of the disease.

Tomato leaf diseases pose a significant threat to global food security, necessitating accurate and efficient detection methods. This paper introduces the Tomato Leaf Disease Visual Language Model (TLDVLM), a novel approach based on the BLIP-2 architecture enhanced with Low-Rank Adaptation (LoRA), for precise classification of 10 distinct tomato leaf diseases. Our methodology integrates a sophisticated image preprocessing pipeline, utilizing GroundingDINO for robust leaf detection and SAM-2 for pixel-level segmentation, ensuring that the model focuses solely on relevant plant tissue. The TLDVLM leverages the powerful multimodal understanding of BLIP-2, with LoRA applied to its Q-Former module, enabling parameter-efficient fine-tuning without compromising performance. Comparative experiments demonstrate that the TLDVLM significantly outperforms baseline models, including CLIP-LoRA and ConvNeXT-tiny, achieving an accuracy of 97.27%, a precision of 0.9587, a recall of 0.9789, and an F1-score of 0.9681. Beyond classification, the finetuned TLDVLM checkpoints are integrated into a practical application for new image inference. This application displays the raw and segmented images, the predicted disease, and offers functionalities to fetch comprehensive information on disease causes and remedies using external APIs (e.g., OpenAI), with an option to download a PDF summary for offline access on a portable device. This research highlights the potential of LoRA-adapted Vision-Language Models in developing highly accurate, efficient, and user-friendly agricultural diagnostic tools.

Plant diseases cause approximately 220 billion USD in annual agricultural losses, driving demand for automated detection systems. This systematic review analyzes deep learning approaches for plant disease detection using RGB and hyperspectral imaging, examining their evolution from classical image processing to modern neural architectures. We evaluate state-of-the-art models across 11 benchmark datasets, revealing significant performance gaps between laboratory conditions (95-99% accuracy) and field deployment (70-85% accuracy). Transformer-based architectures demonstrate superior robustness, with SWIN achieving 88% accuracy on real-world datasets compared to 53% for traditional CNNs. Our analysis identifies three critical deployment constraints: environmental variability sensitivity, economic barriers (500-2000 USD for RGB vs. 20,000-50,000 USD for hyperspectral systems), and interpretability requirements for farmer adoption. Case studies of successful platforms (Plantix with 10+ million users) highlight the importance of offline functionality and multilingual support. We establish evidence-based guidelines prioritizing deployment viability over laboratory optimization and identify key research directions including lightweight model design, cross-geographic generalization, and explainable multimodal fusion. This review provides a comprehensive framework for advancing plant disease detection from research prototypes to practical agricultural tools that can improve global food security.

The application of Fourier transform infrared (FTIR) spectroscopy for non-structural carbohydrates (NSC) prediction as a tool for pre-breeding screening has immense potential but remains to be unexplored, because of technical challenges associated with these measurements. This study investigated the potential of employing FTIR spectroscopy as a high-throughput tool for forecasting NSC content, including total soluble sugar (TSS) and starch content, of 30 rice accessions from the Rice Diversity Panel 1 (RDP1) germplasm and RiceTec hybrids grown in 2019 (320 genotypes) and 2020 cropping (312 genotypes). Partial Least Squares (PLS) regression analysis was used to construct predictive models to estimate NSC content in flag leaves and stem of rice exposed to elevated and ambient nighttime air temperature during the flowering stage of rice. The TSS model exhibited a coefficient of determination (R) value of 0.63 and root mean square error of prediction (RMSEP) values of 3.62 mg g. Notably, the NSC model demonstrated a superior metric performance, with R = 0.66 and RMSEP of 5.58 mg g. The predictive model created in this research effectively measured the NSC composition present in the flag leaves of rice. Expanding the sample size and incorporating additional principal components may enhance the model's predictive accuracy. The FTIR technique can produce fast accurate results and resolve the high analytical costs. Overall, the use of FTIR in conjunction with PLS regression analysis provides a potential tool to advance our understanding of various rice genotypes, particularly concerning their ability to withstand abiotic stress such as HNT.