Accurate detection of strawberry growth stages is vital for optimizing smart agricultural systems, enabling precise resource allocation and effective crop management. This paper proposes SGS-DETR, a lightweight transformer-based model designed for real-time strawberry growth stage detection in greenhouse environments. A total of 2153 images covering all strawberry growth stages were constructed as the experimental dataset, which was split into 1722 training images and 431 testing images at a ratio of 8:2 to ensure objective evaluation. Addressing the increasing demand for intelligent farming solutions capable of handling dense planting patterns and variable environmental conditions, SGS-DETR incorporates advanced deep learning techniques-specifically, a real-time detection transformer-striking a robust balance between detection accuracy and inference speed. The model addresses critical challenges such as elevated false-positive rates and severe occlusion in multi-target greenhouse scenarios prevalent in intelligent agriculture. A comprehensive dataset covering all strawberry growth stages was utilized to train and evaluate the model. Experimental results demonstrate that SGS-DETR achieves high performance across multiple metrics, including a mean average precision (mAP@0.5) of 0.946, an mAP@0.5:0.95 of 0.782, a precision of 0.925, a recall of 0.936, an F1-score of 0.930, and an inference speed of 105.2 FPS, outperforming widely adopted baselines such as Faster R-CNN, the YOLOv series, MobileViT, EfficientViT, and RT-DETR. Furthermore, SGS-DETR maintains high computational efficiency, requiring only 53.9 giga floating-point operations per second (GFLOPs) and a model size of 15.6 MB, thereby ensuring an optimal trade-off between performance and resource utilization. This work offers a promising pathway to enhance the efficiency and sustainability of precision agriculture, strengthen food security, and accelerate the transformation of agricultural practices through AI-driven innovations.

Improving agricultural water-use efficiency under increasing water scarcity is a central challenge for sustainable agriculture. However, intelligent irrigation control is constrained by the high cost of trial-and-error interaction in real fields, while purely reinforcement-learning-based approaches have limited ability to explicitly represent soil water dynamics and crop physiological processes. To address these limitations, this study proposes IrriTwin, a digital-twin-oriented irrigation scheduling framework that couples the Soft Actor-Critic (SAC) algorithm with a soil water model and the GreenLab crop growth model, thereby establishing a model-in-the-loop virtual soil–crop–water environment for policy learning under water resource constraints. Experimental results show that, under the same total irrigation amount, the SAC-based strategy achieves higher fruit biomass than baseline methods. The learned policy exhibits dynamic adaptability to soil moisture conditions and crop phenological stages, effectively balancing vegetative growth and reproductive demands. Additional comparisons with reinforcement learning baselines further indicate that the proposed dual-action SAC achieves more stable convergence and higher final return in the representative maize case. Stability analysis further shows that the strategy remains effective under simulated rainfall and irrigation-failure disturbances. Experiments on sunflower and maize demonstrate the effectiveness of IrriTwin. These results indicate that integrating reinforcement learning with mechanistic soil–crop–water models provides a promising foundation for digital-twin-enabled, precise, and water-efficient irrigation management.

Optimizing plant density through modeling strategies is essential for improving resource-use efficiency and productivity in modern maize production systems. Functional–structural plant models (FSPMs) such as GreenLab provide a powerful framework for simulating crop growth and yield formation, yet their ability to represent interplant competition under high-density conditions remains limited, largely due to the oversimplified treatment of specific leaf area (SLA). To address this gap, a two-year field experiment (2022–2023) composed of four maize plant densities (3, 6, 9 and 12 plants m-2, denoted as PD3, PD6, PD9, and PD12, respectively), was conducted to examine density-driven changes in leaf structural and functional traits and to evaluate the role of SLA plasticity in regulating canopy light and nitrogen use efficiency. Increased plant density substantially reshaped canopy architecture and resource distribution. Compared to PD3, higher densities (PD6, PD9, and PD12) increased leaf orientation value, SLA, photosynthetic nitrogen-use efficiency (PNUE), radiation-use efficiency (RUE) and crop growth rate (CGR), while reducing individual leaf area (LA), specific leaf nitrogen content (SLN) and light-saturated photosynthetic rate (Amax). Further analyses showed that density-induced SLA plasticity enhanced PNUE and adjusted the ratio of nitrogen to light extinction coefficients (KN/KL), thereby partially compensating for the decline in leaf photosynthetic capacity caused by reduced SLN. At the canopy scale, SLA was strongly and positively associated with RUE, highlighting its role in enhancing canopy photosynthesis under interplant competition. Subsequently, the observed SLA responses to plant density were parameterized and incorporated into the GreenLab-Maize model. Compared to the standard model, the revised model markedly improved predictions of LA, leaf area index, RUE and accumulated biomass under high-density conditions. Overall, this study establishes SLA plasticity as a key adaptive trait for resource optimization in dense stands and provide a validated method to enhance the realism of competition simulations in FSPMs.

Forestry knowledge is highly specialized and broad in scopeand is particularly sensitive to regulations and practical standards. To address the knowledge gaps, terminological ambiguities, and factual inaccuracies that general-purpose large language model (LLM) exhibit in forestry scenarios, an integrated framework of "data synthesis-model training-systematic evaluation" was proposed. Based on a general base model, domain-specific instruction fine-tuning was conducted to obtain LinYun, a domain-adapted model for forestry. Experimental results demonstrate that LinYun significantly outperforms general-purpose models of comparable scale in forestry-related tasks, and in some tasks approaches or even surpasses the performance of much larger models.

This article introduces a method for developing an agricultural weed control robot equipped with a delta mechanism. This study focuses on the kinematics and dynamics of Delta mechanisms, the construction of 3D maps for weed detection, and the implementation of control strategies through reinforcement learning (RL). We analyze the forward and backward kinematics of the delta mechanism to link arm length and control output. Our RL module implementation, using proximal policy optimization (PPO), aims to enhance robot weed-removal efficiency. The state space includes the position of the end effector, and the action space is composed of joint angles. The reward function includes multiple factors such as weed control efficiency, crop safety, and energy consumption. Simulations in the V-REP environment validate the control strategy’s effectiveness and its potential application in the real world.

Plant phenomics plays a vital role in smart breeding and precision agriculture, especially under the dual challenges of climate change and food security. This study proposes a low-cost, automated pipeline for 3D reconstruction and leaf segmentation of leafy vegetables and dwarf tomatoes. Using smartphone-captured multi-view videos, we reconstruct high-quality point clouds via SfM and 3D Gaussian Splatting. Backgrounds are removed with HSV filtering and DBSCAN clustering, and spectral clustering enables accurate leaf-level segmentation. Poisson reconstruction is then applied for surface modeling and area calculation. Experiments show that leaf area errors remain within 15%, demonstrating strong accuracy and robustness. The method offers a practical and scalable solution for efficient 3D plant phenotyping.

The development of agriculture faces significant challenges due to population growth, climate change, land depletion, and environmental pollution, threatening global food security [1]. This necessitates the development of sustainable agriculture, where a fundamental step is crop breeding to improve agronomic or economic traits, e.g., increasing yields of crops while decreasing resource usage and minimizing pollution to the environment [2].

Plant simulation models play a significant role in agricultural production and ecological research. However, traditional simulation software is often highly specialized and involves complex parameters, posing substantial entry barriers for non-expert users. With the emergence of large language models (LLMs) and advancements in their function calling capabilities, we propose an interactive simulation system that integrates LLM-based function calling with plant simulation models. This system uses the function-calling capabilities of large models (e.g., DeepSeek-V3) to translate natural language instructions into simulation parameters and commands for the GreenLab platform, thereby enabling interactive control of plant growth models. Furthermore, the system provides real-time visualization of simulation outcomes through a front-end 3D interface. Preliminary results demonstrate that this approach effectively lowers the usability threshold of plant simulation models, allowing non-expert users to conveniently conduct simulations and analyze results using natural language.

Context: Identifying traits that enhance canopy photosynthesis is particularly crucial for sustaining maize productivity under high plant densities. Utilizing a model-assisted approach is an effective strategy to achieve this goal. Objective: The objectives of this study were to (i) integrate a biochemical model of C4 photosynthesis into the existing GreenLab to enhance its capacity for simulating canopy photosynthesis under varying plant densities; (ii) evaluate the model's performance through simulations under different plant densities; and (iii) utilize the model to identify key physiological and structural targets that can enhance productivity under high plant densities. Method: In this study, a two-year field trial of maize (Zea mays L.) was conducted under four plant densities i.e. 3, 6, 9, 12 plants m(-2). Simultaneously, the Functional-Structural Plant Model 'GreenLab' was extended by replacing its existing module for calculating canopy photosynthesis with an update of the C4 photosynthesis model by von Caemmerer (2021). Model parameters (i.e., leaf photosynthesis; sink strength, the capacity of each organ receives biomass; sink variation, each sink strength varies during the duration of organ expansion) governing maize growth and development were estimated using field data collected in 2022. The revised GreenLab was subsequently validated by demonstrating good agreement between independent simulations and experimental observations of maize growth and development across various plant densities in 2023. Results: Leaf photosynthetic and organ sink strength parameters decreased linearly with increasing plant density, while organ sink variation parameters linearly increased. Notably, maximal linear electron transport rate and reproductive organs sink strength and sink variation parameters were quite sensitive to plant density. Modelling trials using only the C4 photosynthetic model revealed that canopy photosynthesis was limited by maximum Rubisco activity, maximal linear electron transport rate, and light distribution under high plant densities. Furthermore, additional modelling studies with the revised GreenLab suggested that synergistically modifying both maximal linear electron transport rate and leaf angle can maximize canopy photosynthesis, thereby improving maize productivity under high plant density. Conclusion: Overall, this study successfully quantified the impact of modifying molecular targets through modelling on enhancing maize canopy photosynthesis under high plant density conditions.

In smart-agriculture, AI is held back by fragmented, hard-to-access datasets. To address this problem, we propose a Decentralized Autonomous Organization (DAO) that couples a consortium Product-Chain (P-Chain) for data custody with a Value-Chain (V-Chain) for decentralized trading. P-Chain registers raw, processed data and model artefacts; V-Chain runs transparent auctions where researchers, farmers, label-service providers and start-ups exchange data, labels or trained models as reusable digital products. Smart-contract pricing and random-validator consensus guarantee authenticity, privacy and fair value flow while preventing collusion. A weed-eradication case study shows that buying ready-made data, labels and models on the platform cuts development cost by 76% compared with in-house collection and training. The architecture thus lowers entry barriers, accelerates AI model iteration, and paves the way for sustainable, data-driven precision agriculture.Copyright (c) 2025 The Authors. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0/)

This article presents an innovative modular planting plant factory inspired by lean manufacturing principle, where planting cells work in a distributed way, coordinated by cloud service. Each planting cell is equipped with controllable light source, CO 2 supplement and fans. All cells are remotely and locally controllable, including instructions on environmental adjustment and planting task scheduling. This system offers flexibility, scalability, and autonomous management of plant factories. Preliminary results indicate its potential in precision agriculture. The modular system adapts to diverse scenarios, focusing on robustness in smart factories and user-friendliness. It envisions urban cultivation as a social platform and aligns with emerging trends in cyber-physical social systems (CPSS) and Distributed Autonomous Organization (DAO). This work represents a convergence of technology, adaptability, and a vision for interactive urban agriculture.

In this digital age, digital twin technology has been widely applied in various fields. However, its use in agriculture is still in the early stages. Existing literature on plant digital twins indicates that they have a limited capacity to interact with real plants or to predict the complete plant growth process. This paper aims to propose a new approach to plant digital twin by combining them with actual plant phenotyping technology. By obtaining plant phenotypic data and integrating them with the plant model, we aim at predicting the entire plant growth process, thereby achieving a dynamic plant digital twin.

Global food market faces escalating risks and uncertainties, bringing great challenges in balancing a country's food supply and demand. Therefore, it is of great significance to carry out crop planning and reasonably divide the planting area of each crop to ensure the national food security. However, the existing planting planning methods have the problems of inaccurate crop price prediction and poor flexibility, and challenges remain on how to motivate farmers. With the rapid development of science and technology, agricultural crop planning techniques have made great progress. This study focuses on agricultural planting planning, exploring both planting area planning based on predicted crop prices and a crop allocation model within a multifarmer context. The regional planting goals are decomposed into specific allocations for individual farmers and plots, addressing objectives including maximizing farmer profits and expanding soybean cultivation for national self-sufficiency. The work employs the long short-term memory (LSTM) model to predict the prices of soybean, wheat, and maize. First, linear programming model is applied to plan planting areas of crops, incorporating constraints to encourage sustainable agricultural practices. Second, a multifarmer crop allocation model, utilizing the double deep Q network (DDQN) algorithm, is developed to enhance the fairness among farmers and assure rotational benefits. Experimental validation confirms the effectiveness of the proposed algorithms, providing valuable decision support for agricultural planning with economic and ecological sustainability.

This study introduces a closed ecological system simulation platform, designed to model the natural ecological cycle within a controlled environment. The paper presents the various types of components within the ecosystem, the methods for mathematical modeling of each component, and the flow of materials between components. It provides a detailed introduction to the design and implementation of the simulation platform, especially the design of the simulation timing sequence. The simulation results from this platform can realistically reflect the overall material flow of the system and its changes over time.