In protected agriculture, accurately identifying the key growth stages of tomatoes plays a significant role in achieving efficient management and high-precision production. However, traditional approaches often face challenges like non-standardized data collection, unbalanced datasets, low recognition efficiency, and limited accuracy. This paper proposes an innovative solution combining generative adversarial networks (GANs) and deep learning techniques to address these challenges. Specifically, the StyleGAN3 model is employed to generate high-quality images of tomato growth stages, effectively augmenting the original dataset with a broader range of images. This augmented dataset is then processed using a Vision Transformer (ViT) model for intelligent recognition of tomato growth stages within a protected agricultural environment. The proposed method was tested on 2723 images, demonstrating that the generated images are nearly indistinguishable from real images. The combined training approach incorporating both generated and original images produced superior recognition results compared to training with only the original images. The validation set achieved an accuracy of 99.6%, while the test set achieved 98.39%, marking improvements of 22.85%, 3.57%, and 3.21% over AlexNet, DenseNet50, and VGG16, respectively. The average detection speed was 9.5 ms. This method provides a highly effective means of identifying tomato growth stages in protected environments and offers valuable insights for improving the efficiency and quality of protected crop production.

Given the constraints of limited cultivated land resources, ensuring and enhancing crop productivity are crucial for food security. This study takes Chengdu as a case study. Using the cultivated land productivity (CLP) evaluation model, we calculated the cultivated land productivity index (CLPI) and analyzed its spatial distribution characteristics. The Geographical Detector model was employed to identify the main factors influencing CLP, and corresponding countermeasures and measures were proposed based on the limiting degrees of these factors. The findings reveal that Chengdu’s CLP index ranges from 1231 to 3053. Global spatial autocorrelation analysis indicates a spatial agglomeration pattern in Chengdu’s overall crop productivity distribution. The local spatial autocorrelation analysis demonstrates that township (street)-level crop productivity in Chengdu is primarily characterized by “high–high”, “low–low”, and “low–high” clusters. Key factors influencing the spatial differentiation of CLP in Chengdu include the agronomic management level, soil bulk density, irrigation guarantee rate, soil body configuration, field slope, and farmland flood control standard. Interaction detection shows that there are both double-factor and nonlinear enhancements among the factors. Specifically, the interaction between soil bulk density and the agronomic management level among other factors have the most explanatory power for the spatial differentiation of CLP. The CLP in Chengdu is highly restricted by its technical level, with the agronomic management level severely limiting CLP by more than 50%. These research results provide a theoretical reference for regional high-standard farmland construction and the protection and utilization of cultivated land resources.

Computer vision has emerged as a critical enabler of sustainable production in protected agriculture by offering efficient and non-invasive crop disease diagnosis. The development of accurate disease recognition models relies heavily on the availability of high-quality image datasets. This study introduces a tomato disease image dataset collected in 2024 from greenhouse facilities within a modern agricultural park in Sichuan Province, China. The dataset comprises 1026 high-resolution images, including 417 images of viral disease, 82 images of gray mold, and 527 images of bacterial wilt, totaling approximately 2.78 GB. Captured under real-world greenhouse conditions and from multiple angles and distances, the images effectively capture multi-scale phenotypic disease features. Manual annotation was conducted using the LabelImg tool under the guidance of plant pathology experts, with labeled regions covering leaves, fruits, and stems. Annotation files are stored in XML format, each corresponding to a specific image. This dataset is well-suited for research in disease classification, object detection, and phenotyping, and supports deep learning model training and cross-crop transfer learning applications.