Accurate yield estimation in oilseed rape (Brassica napus L.) is critical for precision agriculture and breeding selection. Although unmanned aerial vehicle (UAV) remote sensing has become a popular tool for field phenotyping, its application to rapeseed yield estimation remains challenging due to severe canopy occlusion, the small size of siliques, and their complex, elongated morphology. Here, we present a structure-aware rapeseed yield estimation framework that overcomes these limitations by synergizing fine-grained instance segmentation with multi-view oblique photogrammetry. We develop SiliqueSeg, a specialized neural network incorporating self-ensembling attention mechanisms, a tiny detector, and strip pooling to recover fragmented features from high-resolution orthophotos, achieving a mAP50 of 0.957. By decoding plant architectures from multi-view imagery, we establish a mathematical model linking canopy topology to occlusion compensation coefficients. The yield estimation model achieves high accuracy on the training dataset derived from breeding lines, with an R2 of 0.786, an RMSE of 43.95 g/m2, and a mean absolute percentage error (MAPE) of 18.99%. We rigorously validated the effectiveness of the propopsed framework at both the production scale (regional yield mapping) and the breeding scale (genotypic screening). The model maintained robust accuracy across diverse environments, years, and cultivars, enabling regional yield mapping across main rapeseed production regions in Zhejiang Province. Furthermore, GWAS results revealed substantial overlap in candidate genes between predicted and measured yield, successfully identifying 58 known yield-related genes. This scalable framework effectively bridges the gap between canopy traits and actual yield, offering promising implications for precision agriculture and crop breeding.

Organ segmentation of plant point clouds is a prerequisite for the high-resolution and accurate extraction of organ-level phenotypic traits. Although the fast development of deep learning has boosted much research on segmentation of plant point clouds, the existing techniques for organ segmentation still face limitations in resolution, segmentation accuracy, and generalizability across various plant species. In this study, we proposed a novel approach called plant segmentation neural radiance fields (PlantSegNeRF), aiming to directly generate high-precision instance point clouds from multi-view RGB image sequences for a wide range of plant species. PlantSegNeRF performed two-dimensional (2D) instance segmentation on the multi-view images to generate instance masks for each organ with a corresponding instance identification (ID). The multi-view instance IDs corresponding to the same plant organ were then matched and refined using a specially designed instance matching (IM) module. The instance NeRF was developed to render an implicit scene containing color, density, semantic and instance information, which was ultimately converted into high-precision plant instance point clouds based on volume density. The results proved that in semantic segmentation of point clouds, PlantSegNeRF outperformed the commonly used methods, demonstrating an average improvement of 16.1 %, 18.3 %, 17.8 %, and 24.2 % in precision, recall, F1-score, and intersection over union (IoU) compared to the second-best results on structurally complex datasets. More importantly, PlantSegNeRF exhibited significant advantages in instance segmentation. Across all plant datasets, it achieved average improvements of 11.7 %, 38.2%, 32.2% and 25.3 % in mean precision (mPrec), mean recall (mRec), mean coverage (mCov), and mean weighted coverage (mWCov), respectively. Furthermore, PlantSegNeRF demonstrates superior few-shot, cross-species performance, requiring only multi-view images of few plants to train models applicable to specific or similar varieties. This study extends organ-level plant phenotyping and provides a high-throughput way to supply high-quality 3D data for developing large-scale artificial intelligence (AI) models in plant science. (c) 2025 The Authors. Publishing services by Elsevier B.V. on behalf of KeAi Communications Co., Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).

Leaf area index (LAI) is an important structural parameter of crops and it is usually estimated non-destructively using reflectance spectra from various reflectometers. Prevailing models, often trained on single-crop and singleyear data, lack generalizability. As rotation crops with similar morphology, rice and wheat present an opportunity to develop generalized models; however, their spectral response patterns are not well compared, and adaptable multi-year, multi-crop LAI models remain scarce. To bridge this gap, we developed a generalized LAI estimation model for both crops by integrating physically-based simulation with data-driven deep learning. Key steps included canopy spectral simulation, data augmentation, and model construction with a 1D-CNN and transfer learning. The PROSAIL model was employed to simulate canopy reflectance spectra, with crop growth stages stratified into two phenological phases: sowing-heading stage (LAI: 0.01-5, increment: 0.2) and headinggrouting stage (LAI: 3-8, increment: 0.2). To enhance ecological fidelity, the LSMM was integrated to simulate mixed spectral scenarios involving soil background, water interactions, and spike contributions, while 5% Gaussian noise was systematically introduced to approximate real-world environmental variability. The results showed that the R2 values of the SMOTE-1D-CNN model for the different datasets (four rice and two wheat) ranged from 0.62 to 0.87, and the RMSE values ranged from 0.55 to 1.22. The model achieved a relatively high R2 (0.79 f 0.09) for rice LAI estimation but exhibited a larger RMSE (0.8 f 0.29). For wheat, the R2 was slightly lower (0.74 f 0.17), while the RMSE was smaller and more stable (0.56 f 0.01). These discrepancies reflect how crop characteristics or data distribution may influence estimation accuracy. SMOTE is used as a data enhancement to reduce the "high underestimation" phenomenon of the model, and the model performance is stabilized when the multiplicity of the sample size (n) is greater than or equal to 5. And the model input feature importance is only related to the original sample (the original unenhanced dataset) and does not change with "n". This study demonstrates that a hybrid methodology, fusing physically-based simulation with deep learning, offers significant potential for robust, multi-crop LAI inversion, providing novel insights and technical support for crop monitoring and management.

Melon (Cucumis melo L.) is a globally significant horticultural crop, characterized by high nutritional value and substantial commercial status. However, frequent outbreaks of powdery mildew severely threaten its yield and fruit quality. Current early detection methods primarily focus on detached leaf assays, which often lack sufficient model generalization. This study proposes a temporal 3D multispectral point cloud reconstruction method for melon plants by integrating multispectral imaging with 3D reconstruction technology. An Artificial Neural Network (ANN) model for 3D spatial light field distribution was developed based on a hemispherical white reference to achieve precise reflectance calibration of the multispectral point clouds. Post-calibration, the coefficient of variation (CV) for the spectral reflectance of the hemispherical reference in 3D space was reduced to less than 2.4%. On this basis, an early classification model for melon powdery mildew was constructed using Partial Least Squares Discriminant Analysis (PLS-DA) based on the mean reflectance spectra of individual plant point clouds. The results demonstrate that the average recognition accuracy reaches 85.94% from 4 days post-inoculation onwards, enabling disease early warning three days in advance. This research provides critical theoretical support and technical reference for the non-destructive early monitoring and precision smart plant protection of crops in facility agriculture.

With the ongoing advancements in autonomous navigation technology, agricultural robots are increasingly being deployed across various sectors of agriculture. Among the critical components of this technology, dynamic obstacle avoidance in complex agricultural environments serves as the foundation for enhancing the autonomy and safety of these robots. The Dynamic Window Approach (DWA) is a widely recognized method for achieving local obstacle avoidance. It operates by sampling the robot's velocity space and then evaluating the sampled trajectories using a value function to determine the optimal velocity pair. However, a significant limitation of the traditional DWA method lies in its fixed weights for the value function, which restricts its performance to manual tuning and renders it less adaptable to intricate and dynamic obstacle environments. To address this limitation, we introduced an innovative approach by integrating the Twin Delayed Deep Deterministic Policy Gradient (TD3) method into the weight determination process of the DWA algorithm's value function. This integration enabled the weight coefficients to adapt dynamically in response to environmental variations, thereby enhancing the algorithm's flexibility and effectiveness. Our extensive simulation and field testing revealed that while the traditional DWA algorithm struggled to navigate complex dynamic obstacle environments, the proposed TD3DWA algorithm achieved a success rate of over 90 % in obstacle avoidance. This outcome underscored the algorithm's adaptability and robustness, positioning it as a reliable solution for ensuring safe and efficient navigation in agricultural robotics.

Quantitative characterization of complete canopy architecture is essential for accurate evaluation of crop photosynthesis and yield potential, thereby supporting crop ideotype design. Although various sensing technologies enable three-dimensional (3D) reconstruction of individual plants and canopies, they often fail to describe canopy architecture accurately because of severe occlusion in dense populations. To address this limitation, we developed an effective framework for the 3D reconstruction of complex and dynamic population-scale canopy architecture in rapeseed using unmanned aerial vehicle multi-view imagery combined with a novel point cloud completion model. A complete point cloud generation pipeline was first established to enable automated training data annotation, allowing discrimination between surface points and occluded points within the canopy. The proposed crop population point cloud completion network (CP-PCN) integrates a multi-resolution dynamic graph convolutional encoder, a point pyramid decoder, a dynamic graph convolutional feature extractor, and a generative adversarial network-based loss function to predict occluded canopy points. CP-PCN achieved chamfer distance values of 3.35 to 4.51 cm across four growth stages, outperforming the state-of-the-art transformer-based method PoinTr. Ablation analyses confirmed that each of the four modules contributes to overall model accuracy. In addition, validation experiments showed that the improved architectural completeness achieved by CP-PCN resulted in more accurate yield estimation compared with incomplete and PoinTr-completed point clouds. CP-PCN also demonstrated strong cross-crop generalizability by successfully reconstructing mature rice canopies. Overall, this framework provides a scalable approach for quantitative analysis of complex canopy architectures in field-grown crops.

Accurate estimation of comprehensive traits such as yield and quality is crucial for optimizing agricultural management practices across the tomato industry chain. Traditional manual methods are time-consuming, laborintensive, and prone to errors, reducing estimation accuracy. In contrast, modern intelligent estimation approaches based on multi-temporal spatial and spectral feature fusion offer improved efficiency and accuracy but still face challenges such as non-generalizable segmentation models, asynchronous feature extraction and weak correlations. This study proposes a novel pipeline for estimating yield and quality of greenhouse tomatoes using temporal semantic multispectral (TSM) point clouds. An unsupervised deep learning model was designed to register RGB-D images and multispectral (MS) images collected by an unmanned ground vehicle (UGV) plant phenotyping platform. The digital number (DN) point clouds of tomato organs were reconstructed based on the masks predicted by SegFormer with fusion of multispectral and depth modalities (MSD-SF). These point clouds were then radiometrically calibrated using neural reference field with sparse viewpoints (NeREF-S) to generate accurate reflectance point clouds. Finally, multi-temporal spatial-spectral features of tomatoes were extracted from the TSM point clouds, and random forest regression models were developed to estimate traits such as fruit flavor preference, water content, brix, acidity, brix-to-acid ratio, vitamin C content, single-fruit mass, and singleplant yield. The image registration model achieved high accuracy on the test set, with average structural similarity index measure, peak signal-to-noise ratio and learned perceptual image patch similarity of 0.238, 13.116 dB, and 0.374, respectively. The MS point clouds calibrated by NeREF-S significantly improved the signal-tonoise ratio to 11.56 dB. The average rRMSE for all trait estimations was 9.03 %. The results indicate that the proposed estimation method is efficient and accurate, holding promise to become a new paradigm for estimating the comprehensive traits of greenhouse tomatoes.

Color accuracy and consistency in remote sensing imagery are crucial for reliable plant health monitoring, precise growth stage identification, and stress detection. However, without effective color correction, variations in lighting and sensor sensitivity often cause color distortions between images, compromising data quality and analysis. This study introduces a novel in-flight color correction approach for RGB imagery using cooperative dual unmanned aerial vehicle (UAV) flights integrated with a color chart (CoF-CC). The method employs a master UAV equipped with an RGB camera for image acquisition and a synchronized secondary UAV carrying a ColorChecker (X-Rite) chart, ensuring persistent visibility of the chart within the imaging field of the master UAV for the calculation of a color correction matrix (CCM) for in-flight image correction. Field experiments validated the method by analyzing cross-sensor color consistency, assessing color measurement accuracy on field-grown rice leaves, and demonstrating its practical applications using rice maturity estimation as an example. The results indicated that the CCM significantly enhanced color accuracy, with a 66.1 ​% reduction in the average CIE 2000 color difference (ΔE), and improved color consistency among the six RGB sensors, with a 70.2 ​% increase in the intracluster distance. CoF-CC subsequently reduced ΔE from 18.2 to 5.0 between the corrected rice leaf color and ground-truth measurements, indicating that the color differences were nearly perceptible to the human eye. Moreover, the corrected imagery significantly enhanced the rice maturity prediction accuracy, improving the R2 from 0.28 to 0.67. In summary, the CoF-CC method standardizes RGB images across diverse lighting conditions and sensors, demonstrating robust performance in color analysis and interpretation under open-field conditions.

Quantitative descriptions of the complete canopy architecture are essential for accurately evaluating crop photosynthesis and yield performance to guide ideotype design. Although various sensing technologies have been developed for three-dimensional (3D) reconstruction of individual plants and canopies, they failed to obtain an accurate description of canopy architectures due to severe occlusion among complex canopy architectures. We proposed an effective method for 3D reconstruction of complex, dynamic population canopy architecture for rapeseed crops with a novel point cloud completion model. A complete point cloud generation framework was developed for automated annotation of the training dataset by distinguishing surface points from occluded points within canopies. The crop population point cloud completion network (CP-PCN) was then designed with a multi-resolution dynamic graph convolutional encoder (MRDG) and a point pyramid decoder (PPD) to predict occluded points. To further enhance feature extraction, a dynamic graph convolutional feature extractor (DGCFE) module was proposed to capture structural variations over the whole rapeseed growth period. The results demonstrated that CP-PCN achieved chamfer distance (CD) values of 3.35 cm -4.51 cm over four growth stages, outperforming the state-of-the-art transformer-based method (PoinTr). Ablation studies confirmed the effectiveness of the MRDG and DGCFE modules. Moreover, the validation experiment demonstrated that the silique efficiency index developed from CP-PCN improved the overall accuracy of rapeseed yield prediction by 11.2

Biomass estimation of oilseed rape is crucial for optimizing crop productivity and breeding strategies. While UAV-based imaging has advanced high-throughput phenotyping, current methods often rely on orthophoto images, which struggle with overlapping leaves and incomplete structural information in complex field environments. This study integrates 3D Gaussian Splatting (3DGS) with the Segment Anything Model (SAM) for precise 3D reconstruction and biomass estimation of oilseed rape. UAV multi-view oblique images from 36 angles were used to perform 3D reconstruction, with the SAM module enhancing point cloud segmentation. The segmented point clouds were then converted into point cloud volumes, which were fitted to ground-measured biomass using linear regression. The results showed that 3DGS (7 k and 30 k iterations) provided high accuracy, with peak signal-to-noise ratios (PSNR) of 27.43 and 29.53 and training times of 7 and 49 min, respectively. This performance exceeded that of structure from motion (SfM) and mipmap Neural Radiance Fields (Mip-NeRF), demonstrating superior efficiency. The SAM module achieved high segmentation accuracy, with a mean intersection over union (mIoU) of 0.961 and an F1-score of 0.980. Additionally, a comparison of biomass extraction models found the point cloud volume model to be the most accurate, with an determination coefficient (R2) of 0.976, root mean square error (RMSE) of 2.92 g/plant, and mean absolute percentage error (MAPE) of 6.81 %, outperforming both the plot crop volume and individual crop volume models. This study highlights the potential of combining 3DGS with multi-view UAV imaging for improved biomass phenotyping.

Despite the success of deep convolutional neural networks (DCNNs) in various applications, optimizing them for specific tasks remains challenging due to the complex manual tuning of hyperparameters. This approach is often ineffective when balancing multiple objectives, as it relies heavily on trial and error. This study proposes an innovative multi-objective genetic algorithm (MOGA) approach to automatically learn highly efficient and resource-saving DCNN architectures, in short, termed as MOGA-DCNN, for a given plant image segmentation task. To this end, a novel encoding technique was proposed to simplify the structure of the candidate solutions and constrain the search space in such a way that a Pareto set of non-dominated solutions can be explored efficiently through genetic operators, reducing computational complexity, and improving performance. We have evaluated this approach on different datasets collected from fruit trees and oilseed rape crops under controlled and uncontrolled conditions. The results demonstrated the capability of the proposed MOGA-DCNN to automatically construct variable-length DCNN architectures for each dataset. The storage size of the architecture parameters (71 K) was only 0.24 % of the well-known SegNet. The evolved model classifies an image 15 - 18 times faster than DeepLab v3+, indicating an overwhelming advantage in image segmentation. These results suggest the prospect of model transferability to different image segmentation tasks, and it could be integrated into embedded system devices with an extremely low computational cost.

Large-scale manual measurements of plant architectural traits in tomato growth are laborious and subjective, hindering deeper understanding of temporal variations in gene expression heterogeneity. This study develops a high-throughput approach for characterizing tomato architectural traits at different growth stages and mapping temporal broad-sense heritability using an unmanned ground vehicle-based plant phenotyping system. The SegFormer with fusion of multispectral and depth imaging modalities was employed to semantically segment plant organs from the registered RGB-D and multispectral images. Organ point clouds were then generated and clustered into instances. Finally, six key architectural traits, including fruit spacing (FS), inflorescence height (IH), stem thickness (ST), leaf spacing (LS), total leaf area (TLA), and leaf inclination angle (LIA) were extracted and the temporal broad-sense heritability folds were plotted. The root mean square errors (RMSEs) of the estimated FS, IH, ST, and LS were 0.014, 0.043, 0.003, and 0.015 m, respectively. The visualizations of the estimated TLA and LIA matched the actual growth trends. The broad-sense heritability of the extracted traits exhibited different trends across the growth stages: (i) ST, IH, and FS had a gradually increased broad-sense heritability over time, (ii) LS and LIA had a decreasing trend, and (iii) TLA showed fluctuations (i.e. an M-shaped pattern) of the broad-sense heritability throughout the growth period. The developed system and analytical approach are promising tools for accurate and rapid characterization of spatiotemporal changes of tomato plant architecture in controlled environments, laying the foundation for efficient crop breeding and precision production management in the future.

Deep learning-based crop and weed detection is essential for modern precision weed control. But its effectiveness is limited when facing newly presented weed species due to the impracticality of collecting large, balanced training datasets in field conditions. To address these challenges, this study presents a few-shot learning framework that achieves rapid and effective adaptation to new weed species by leveraging domain-specific characteristics of plant detection. We proposed few-shot enhanced attention (FSEA) network, built upon Faster R-CNN, which implements three prior knowledge in weed detection through: (1) designing a channel attention-based feature fusion module with an excess-green feature extractor to leverage color characteristics of plants and background, (2) designing a feature enhancement module to accommodate diverse plant morphologies, and (3) applying an optimized loss function designed specifically for plant occlusion scenarios. Using commonly observed crop and weed species (common beet, sugarcane, barnyard grass, field pennycress and Chinese money plant) as base classes, FSEA achieved an all-class mAP of 0.416 and a novel-class mAP of 0.346 when adapting to less frequent weed species (common purslane, Asian copperleaf, goosefoot, clover, and goosegrass), after training for 40 epochs using only 30 samples per species. This performance significantly outperforms state-of-the-art few-shot detectors (TFA, FSCE, Meta R-CNN, Meta-DETR, DCFS, DiGEO) and traditional detector YOLOv7, indicating the effectiveness of incorporating domain-specific prior knowledge into few-shot weed detection. This study provides a fundamental methodology for rapid adaptation of weed detection systems to new environments and species, making automated weed management more practical and accessible for various agricultural applications. The source code and dataset are publicly available (https://github.co m/skyofyao/FSEA) to facilitate further research in this domain.

Remote sensing images often suffer from color distortion, which can pose significant challenges for accurate data interpretation. To overcome this issue, this study developed a new approach called the multichannel convolutional neural network (mCNN). This technique treated color calibration of remote sensing images as an image-to-image (im2im) regression problem and used a neural network to learn a mapping function between the distorted and calibrated images by assigning varying weights to pixels across the scene. The mCNN consisted of three convolutional layer groups, each applied to one image channel separately. The input to each group was a residual channel, which is the difference between the distorted and calibrated images for that channel. A fusion layer was added to concatenate the outputs of the last convolutional layer from all channels, and then applied a regression layer to generate a full output image while preserving its structural details. Unlike other deep learning methods that work in multiple steps, the mCNN performed color calibration in a single step, leading to more efficient processing of large datasets. The method was validated using high-quality multitype ground-truth datasets and compared with other color correction methods as well as its variants. Results showed that the mCNN model outperformed all competitors in terms of the mean value of the estimation errors, with a margin of 34.24%. These findings suggest that the proposed method is highly effective in addressing color distortion issues in remote sensing images, which is critical for accurately representing vegetation properties.

The number of panicles per unit area (PNpA) is one of the key factors contributing to the grain yield of rice crops. Accurate PNpA quantification is vital for breeding high-yield rice cultivars. Previous studies were based on proximal sensing with fixed observation platforms or unmanned aerial vehicles (UAVs). The near-canopy images produced in these studies suffer from inefficiency and complex image processing pipelines that require manual image cropping and annotation. This study aims to develop an automated, high-throughput UAV imagery-based approach for field plot segmentation and panicle number quantification, along with a novel classification method for different panicle types, enhancing PNpA quantification at the plot level. RGB images of the rice canopy were efficiently captured at an altitude of 15 m, followed by image stitching and plot boundary recognition via a mask region-based convolutional neural network (Mask R-CNN). The images were then segmented into plot-scale subgraphs, which were categorized into 3 growth stages. The panicle vision transformer (Panicle-ViT), which integrates a multipath vision transformer and replaces the Mask R-CNN backbone, accurately detects panicles. Additionally, the Res2Net50 architecture classified panicle types with 4 angles of 0°, 15°, 45°, and 90°. The results confirm that the performance of Plot-Seg is comparable to that of manual segmentation. Panicle-ViT outperforms the traditional Mask R-CNN across all the datasets, with the average precision at 50% intersection over union (AP50) improved by 3.5% to 20.5%. The PNpA quantification for the full dataset achieved superior performance, with a coefficient of determination (R2) of 0.73 and a root mean square error (RMSE) of 28.3, and the overall panicle classification accuracy reached 94.8%. The proposed approach enhances operational efficiency and automates the process from plot cropping to PNpA prediction, which is promising for accelerating the selection of desired traits in rice breeding.

With the growing prominence of UAV-based low-altitude remote sensing in agriculture, the acquisition and processing of high-quality UAV remote sensing images is paramount. The purpose of this study is to investigate the impact of various parameter settings on image quality and optimize these parameters for UAV operations to enhance efficiency and image quality. The study examined the effects of three parameter settings (exposure time, flight altitudes and forward overlap (OF)) on image quality and assessed images obtained under various conditions using signal-to-noise ratio (SNR) and BRISQUE algorithms. The results indicate that the setting of exposure time during UAV image acquisition directly affects image quality, with shorter exposure times resulting in lower SNR. The optimal exposure times for the RGB and MS cameras have been determined as 0.8 ms to 1.1 ms and 4 ms to 16 ms, respectively. Additionally, the best image quality is observed at flight altitudes between 15 and 35 m. The setting of UAV OF complements exposure time and flight altitude; to ensure the completeness of image acquisition, it is suggested that the flight OF is set to approximately 75% at a flight altitude of 25 m. Finally, the proposed image redundancy removal method has been demonstrated as a feasible approach for reducing image mosaicking time (by 84%) and enhancing the quality of stitched images (by 14%). This research has the potential to reduce flight costs, improve image quality, and significantly enhance agricultural production efficiency.

Advancements in unmanned aerial vehicle (UAV) remote sensing with spectral imaging enable efficient assessment of critical agronomic traits. However, existing reflectance calibration or generation methods suffer from limited prediction accuracy and practical flexibility. This study explores reliable and cost-efficient methods for the accurate conversion of digital number values acquired from a multispectral imager into reflectance, leveraging real-time solar spectra as references. To ensure consistent measurements of incident light, an upward gimbal-mounted downwelling spectrometer was attached to the UAV, and a sinusoidal model was developed to correct for solar position variability. Using principal component analysis on the reference solar spectrum for band selection, a multiple linear regression model with four sensitive bands (4-Band MLR) and a 30 nm bandwidth achieved performance comparable to the direct correction method. The root mean square error (RMSE) for reflectance prediction improved by 86.1 fluctuating cloudy conditions and by 59.6 sensor method averaged across different weather conditions. The RMSE was calculated as 2.24 campaign conducted at various times throughout a sunny day. Implementing the 4-Band MLR model enhanced the consistency of canopy reflectance within a homogeneous vegetation area by 95.0 field under significant cloud fluctuations. Additionally, improvements of 86.0 and 90.3 vegetation index (NDVI; a ratio index) and the difference vegetation index (DVI; a non-ratio index), respectively.

Remote sensing-based techniques have been widely used for chlorophyll content (Cab) estimations, while they are challenging when transferred across different species. Sun-induced chlorophyll fluorescence (SIF) provides a new approach to address these issues. This research explores whether SIF has transferability for Cab estimation and to enhance between-species transferability. Here, three rice datasets and a rapeseed dataset were collected. Initially, direct transfer models were constructed using partial least squares regression (PLSR) based on SIF yield (SIFY) and reflectance, respectively. Subsequently, methods were employed within the rice datasets to improve the models’ transferability. Finally, the between-species transferability of two data sources was validated in the rapeseed dataset. Direct transfer models indicated that the reflectance-based model exhibited a higher accuracy in predicting Cab when the training dataset acquired sufficient features, whereas the SIFY-based model showed better performance with fewer features. Spectral preprocessing methods can enhance the transferability, especially for SIFY-based models. In addition, supplementing 10% of out-of-sample data significantly improved the transferability. The proposed methods only require a small amount of new data to extend the original model for predicting Cab in other species. Specifically, the new method reduced the average RMSE based on SIFY and reflectance models by 23.59% and 35.51%, respectively.