Search Results (3)

The axillary buds that grow between the main and lateral branches of tomato plants waste nutrients and lead to a decrease in yield, necessitating regular removal. Currently, these buds are removed manually, which requires substantial manpower and incurs high production costs, particularly on a large scale. Replacing manual labor with robots can lead to cost reduction. However, a critical challenge is the accurate multi-target identification of tomato plants and precise positioning for axillary bud removal. Therefore, this paper proposes a multi-target identification and localization method for tomato plants based on the VGG16-UNet model. The average intersection and pixel accuracies of the VGG16-UNet model after introducing the pretrained weights were 85.33% and 92.47%, respectively, which were 5.02% and 4.08% higher than those of the VGG16-UNet without pretrained weights, achieving the identification of main branches, side branches, and axillary bud regions. Then, based on the multi-objective segmentation of the tomato plants in the VGG16-UNet model, the regions of the axillary buds in the tomato plants were identified by HSV color space conversion and color threshold range selection. Morphological dilation and erosion operations were used to remove noise and connect adjacent regions of the same target. The endpoints and centroids of the axillary buds were identified using the feature point extraction algorithm. The left and right positions of the axillary buds were judged by the relationship between the position of the axillary bud centroid and the position of the main branch. Finally, the coordinate parameters of the axillary bud removal points were calculated using the feature points to determine the relationship between the position of the axillary bud and the position of the branch. Experimental results showed that the average accuracy of the axillary bud pruning point recognition was 85.5%. Full article

In this study, we improve the efficiency of automated tomato harvesting by integrating deep learning into state-of-the-art image processing techniques, which improves the accuracy and efficiency of detection algorithms for robotic systems. We develop a hybrid model that combines convolutional neural networks’ dual two-dimensional matrices for classification and part affinity fields. We use data augmentation to improve the robustness of the model and reduce overfitting. Additionally, we apply transfer learning to solve the challenging problem of improving the accuracy of identifying a tomato’s center of gravity. When tested on 2260 diverse images, our model achieved a recognition accuracy of 96.4%, thus significantly outperforming existing algorithms. This high accuracy, which is specific to the environmental conditions and tomato varieties used, demonstrates the adaptability of the model to real-world agricultural conditions. Our results represent a significant advancement in the field of agricultural autotomization by demonstrating an algorithm that not only identifies ripe tomatoes for robotic harvesting with high accuracy, but also adapts to various agricultural conditions. This algorithm should reduce manual labor in agriculture and offer a more efficient and scalable approach for the future agricultural industry. Full article

Recognition and localization of fruits are key components to achieve automated fruit picking. However, current neural-network-based fruit recognition algorithms have disadvantages such as high complexity. Traditional stereo matching algorithms also have low accuracy. To solve these problems, this study targeting greenhouse tomatoes proposed an algorithm framework based on YOLO-TomatoSeg, a lightweight tomato instance segmentation model improved from YOLOv5n-seg, and an accurate tomato localization approach using RAFT-Stereo disparity estimation and least squares point cloud fitting. First, binocular tomato images were captured using a binocular camera system. The left image was processed by YOLO-TomatoSeg to segment tomato instances and generate masks. Concurrently, RAFT-Stereo estimated image disparity for computing the original depth point cloud. Then, the point cloud was clipped by tomato masks to isolate tomato point clouds, which were further preprocessed. Finally, a least squares sphere fitting method estimated the 3D centroid co-ordinates and radii of tomatoes by fitting the tomato point clouds to spherical models. The experimental results showed that, in the tomato instance segmentation stage, the YOLO-TomatoSeg model replaced the Backbone network of YOLOv5n-seg with the building blocks of ShuffleNetV2 and incorporated an SE attention module, which reduced model complexity while improving model segmentation accuracy. Ultimately, the YOLO-TomatoSeg model achieved an AP of 99.01% with a size of only 2.52 MB, significantly outperforming mainstream instance segmentation models such as Mask R-CNN (98.30% AP) and YOLACT (96.49% AP). The model size was reduced by 68.3% compared to the original YOLOv5n-seg model. In the tomato localization stage, at the range of 280 mm to 480 mm, the average error of the tomato centroid localization was affected by occlusion and sunlight conditions. The maximum average localization error was ±5.0 mm, meeting the localization accuracy requirements of the tomato-picking robots. This study developed a lightweight tomato instance segmentation model and achieved accurate localization of tomato, which can facilitate research, development, and application of fruit-picking robots. Full article