Deep learning-based methods have proven to be effective for various purposes in the agricultural sector. However, these methods require large amounts of labelled data, which are difficult to prepare and preprocess. To overcome this problem, we propose the use of a latent diffusion model for plant disease detection by generating unseen class images. In this study, we used images of healthy and diseased grape leaves as training datasets and utilized the latent diffusion model, known for its superior performance in image generation, to generate images of diseased apple leaves that were not included in this dataset. Image-to-image generation was utilized to preserve the original healthy leaf features, which enabled the appropriate image generation of diseased apple leaves. To ascertain whether the generated diseased apple leaf images could be used to detect leaf diseases, a deep learning-based classification model was trained to discriminate between diseased and healthy apple leaves from a dataset with a mixture of actual and generated images. Results showed that leaves were accurately classified, indicating that diseased apple leaves not included in the training data could be used to identify the actual diseased apple leaves. Our approach opens up new avenues for improving plant disease detection methods.

This study investigated the interoperability of a tomato fruit detection model trained using nighttime images from two greenhouses. The goal was to evaluate the performance of the models in different environmets, including different facilities, cultivation methods, and imaging times. An innovative imaging approach is introduced to eliminate the background, highlight the target plants, and test the adaptability of the model under diverse conditions. The results demonstrate that the tomato fruit detection accuracy improves when the domain of the training dataset contains the test environment. The quantitative results showed high interoperability, achieving an average accuracy (AP₅₀) of 0.973 in the same greenhouse and a stable performance of 0.962 in another greenhouse. The imaging approach controlled the lighting conditions, effectively eliminating the domain-shift problem. However, training on a dataset with low diversity or inferring plant appearance images but not on the training dataset decreased the average accuracy to approximately 0.80, revealing the need for new approaches to overcome fruit occlusion. Importantly, these findings have practical implications for the application of automated tomato fruit set monitoring systems in greenhouses to enhance agricultural efficiency and productivity.

Early fruit size prediction in greenhouse tomato (Solanum lycopersicum L.) is crucial for growers managing cultivars to reduce the yield ratio of small-sized fruit and for stakeholders in the horticultural supply chain. We aimed to develop a method for early prediction of tomato fruit size at harvest with machine learning algorithm, and three machine learning models (Ridge Regression, Extra Tree Regrreion, CatBoost Regression) were compared using the PyCaret package for Python. For constructing the models, the fruit weight estimated from the fruit diameter obtained over time for each cumulative temperature after anthesis was used as explanatory variable and the fruit weight at harvest was used as objective variable. Datasets for two different prediction periods after anthesis of three tomato cultivars (“CF Momotaro York,” “Zayda,” and “Adventure.”) were used to develop tomato size prediction models, and their performance was evaluated. We also aimed to improve the model adding the average temperature during the prediction period as an explanatory variable. When the estimated fruit size data at cumulative temperatures of 200°C d, 300°C d, and 500°C d after anthesis were used as explanatory variables, the mean absolute percentage error (MAPE) was lowest for “Zayda,” a cultivar with stable fruit diameter, at 9.8% for Ridge Regression. When the estimated fruit size at cumulative temperatures of 300°C d, 500°C d, and 800°C d after anthesis were used as explanatory variables for Ridge Regression, the MAPE decreased for all cultivars: 10.1% for “CF Momotaro York,” 8.8% for “Zayda,” and 10.0% for “Adventure.” In addition, incorporating the average temperature during the fruit size prediction period as an explanatory variable slightly increased model performance. These results indicate that this method could effectively predict tomato size at harvest in three cultivars. If fruit diameter data acquisition could be automated or simplified, it would assist in cultivation management, such as tomato thinning.

Strawberries are an economically valuable fruit in Japan; hence, their production must be maintained throughout the year. In this study, we investigated the effect of temperature on the number of days required for each fruit growth and ripening stage in order to obtain a basic knowledge for the...

The scale of horticultural facilities in Japan is expanding, making the efficient management of labor costs essential, particularly in large-scale tomato production. This study developed a consistent and practical system for predicting harvest working time and estimating the quantity and weight of harvested fruit using panoramic images of cultivation rows. The system integrates a deep learning model, the Mask ResNet-50 convolutional neural network, to count harvestable fruits from images and a predictive algorithm to estimate working time based on the fruit count. The results indicated that the average for all workers could be predicted with an error margin of 30.1% when predicted three days before the harvest date and 15.6% when predicted on the harvest date. The trial also revealed that the accuracy of the predictions varied based on workers’ experience and cultivation methods. This study highlights the system’s potential to optimize harvesting plans and labor allocation, providing a novel tool for reducing labor costs while maintaining efficiency in large-scale tomato greenhouse production.

The objective of this study is to develop an AI-based fruit monitoring system to address the issue of image background noise when automatically counting tomatoes grown in greenhouse horticulture. The system consists of a scanning device and a rail-guided vehicle. This system creates a panoramic image of the crop canopy and automatically counts the number of fruits using deep learning-based fruit detection model (Mask R-CNN). A focused illumination unit on our scanning device creates a light-dark contrast between the near and far sides of the camera, removing unnecessary objects from the background of the images. We applied our system to detect tomato fruits grown in an experimental greenhouse and successfully achieved good performance (AP50 = 0.95).