Tissue culture rapid propagation cannot be reduced the production that is caused by environmental factors, compared with traditional flower cultivation. Since phalaenopsis flowers are very popular for their unique appearance, tissue culture rapid propagation can be expected to cultivate phalaenopsis for the high cultivation speed in industrial production. Current phalaenopsis rapid propagation cannot fully meet the large-scale production in recent years, due to the high repetition and labor consumption. Therefore, it is in high demand for automated phalaenopsis tissue-cultured seedlings' rapid propagation. Among them, image recognition can be used to detect the picking points of seedlings in the process of the robot arms. However, current algorithms of image recognition are only applicable to the specific growth stages, due to the differences in the image characteristics of seedlings at different stages. This study aims to improve the adaptability and efficiency of seedling grasping point detection. A seedling grasping point localization was also proposed using the improved U²-Net salient detection network (MBU²-Net⁺). A seedling grasping end effector was then designed. The overall steps of seedling grasping included visual detection, grasping point localization, and a robotic arm control module. Firstly, industrial cameras were used to capture the images of seedlings. The salient map of the seedlings was obtained after the MBU²-Net⁺ salient detection network. Secondly, the salient map was processed by filtering, while the skeleton extraction was to extract the intersection points of the skeleton lines. Then, the intersection points were clustered using K Nearest Neighbours (KNN) to remove the salient outliers. The grasping angle was obtained to locate the seedling grasping point using morphological analysis after fitting the line using Random Sample Consensus (RANSAC). Finally, the grasping point location data was sent to the robotic arm for grasping. The salient detection experiment was carried out to compare MBU2-Net⁺ with Res2Net-PoolNet, U²-Net, and U²-Net⁺. The average absolute error of MBU²-Net⁺ was 0.002, the maximum F₁ score was 0.993, the FPS was 33.99, and the model weight size was 2.37MB. All of these metrics were optimal. The success rate of grasping point detection of 112 seedlings in four groups was 85.71% in the grasping point of seedling detection experiment using MBU²-Net⁺. The failure of grasping point extraction was attributed to the complex structure of certain seedlings, indicating a significant number of mature leaves and roots. Consequently, the intersections in the obtained skeleton image deviated significantly from the actual locations of roots or leaves, leading to grasping points out of the appropriate regions. Six seedlings were selected to verify the adaptability from the mother bottle to the intermediate bottle stage, three seedlings from the intermediate bottle to the daughter bottle stage, and two virtual seedlings in the rapid propagation process of phalaenopsis. The overall success rate was 81.82%. A high success rate was achieved to extract the grasping points for the orchid seedlings in the entire process of orchid tissue culture. Furthermore, a certain level of generalization was suitable for the other types of seedlings with similar structures. In summary, the capability of the improved model can be expected to effectively identify the grasping points for the orchid seedlings in the entire process of orchid tissue culture, even with the limited datasets. The adaptability and efficiency were significantly enhanced to detect the grasping points for new seedlings. The findings can serve as a technical reference in the development of automated and rapid propagation techniques for tissue-cultured seedlings.

Phalaenopsis is renowned for its butterfly-like flowers. It is highly favored and holds promising market prospects at present. The rapid propagation of phalaenopsis primarily relies on tissue culture technology. However, this process is characterized by strong dependence on manual labor, highly repetitive operations, and low production efficiency. Therefore, it is imperative to automate the tissue culture of phalaenopsis. It is also crucial to accurately detect the gripping point for the transplantation operation in tissue culture. Particularly, the mask center can better represent the gripping point compared with the bounding box center. In this study, an improved YOLOv11n-seg-based segmentation model was proposed for the tissue-cultured seedling, in order to improve the segmentation accuracy of the gripping area and facilitate deployment by reducing the computational complexity. Firstly, the backbone feature extraction network of the original model was replaced with the RepViT network. Its reparameterization structure and modular decoupling enhanced feature extraction while lowering complexity. Secondly, a lightweight cross-scale feature fusion module (CCFM) was introduced into the neck network of the original model. The progressive fusion of adjacent scales was achieved through the layer-by-layer stacked fusion blocks. The CCFM effectively avoided the high computational overhead and improved the model's detection capability for small and medium-sized targets. Meanwhile, to improve the annotation efficiency on the tissue-cultured seedling segmentation dataset, an automatic annotation method based on Grounded SAM 2 was proposed. Specifically, the outputs of Grounded SAM 2 were processed using AddSub algorithm, including dynamic area threshold denoising, morphological operations, and mask difference fusion operations. A series of experiments were conducted on a phalaenopsis tissue-cultured seedling image dataset consisting of 1000 training images and 172 validation images. Experimental results showed that the improved model can accurately locate the gripping point of the tissue-cultured seedling, with an average Euclidean distance of only 1.95 mm for the mask centroid of the gripping region between the model and the manually annotated. The improved model achieved precision, recall, mAP₅₀, and mAP_50:95 of 96.0%, 81.8%, 87.7%, and 67.2%, respectively, indicating the improvement of 0.6, 2.8, 3.3, and 8.5 percentage points, compared with the original model YOLOv11n-seg. The parameter and FLOPs were reduced by 46.7% and 20.6%, respectively, and the size of the improved model was only 3.8 MB,achieving model lightweighting while enhancing segmentation accuracy. The automatic annotation was successfully realized on 1172 tissue-cultured seedling images. A high level of consistency was found with the manual annotation from Labelme and ISAT, with an annotation success rate of 84.5%, fully meeting the practical requirement. Additionally, the average annotation time per image was 6.6 s, which was 176.6 and 50.0 s faster than manual annotation using Labelme and ISAT, respectively. The efficiency of building the training dataset was significantly improved for the phalaenopsis tissue-cultured seedling. The gripping point was effectively extracted for the tissue-cultured seedling, and the tissue-cultured seedling images can be automatically annotated. The finding can provide valuable technical reference for the automation of the tissue culture.

Phalaenopsis has been widely popular ornamental plants, due to its unique appearance and excellent market prospects. The commonly used breeding for phalaenopsis is tissue culture at the same time. However, this task involves high repetition and labor intensity. It is still challenging to increase the production of butterfly orchids. This study aims to improve the yield of tissue culture for the labor-saving in phalaenopsis production. A visual servo platform was proposed with the robotic arm gripping. The picking and planting of tissue-cultured seedlings were realized on the conveyor belt. The visual robotic arm was constructed as a 6-degree-of-freedom robotic arm with eyes on the hand, according to the uncalibrated visual servo control. Firstly, the images were captured from the camera. Image detection was implemented to obtain the image coordinates of the roots, stems, and leaves of phalaenopsis tissue-cultured seedlings. Then, the image data was analyzed to determine the angle between the tissue-cultured seedlings and the end clamp of the robotic arm, as well as the gripping point. According to this angle, the end posture of the robotic arm was obtained during gripping. Afterward, the difference in picking point coordinates was calculated between the current and expected image for the tissue-cultured seedlings. The obtained data was then fed into the fuzzy as an input. The acceleration value was derived to control the speed at the end of the robotic arm from the input quantity. The end gripper of the robotic arm was used to track the tissue-cultured seedlings during planting. The acceleration control was utilized to remain relatively stationary in the planting experiment. The tracking speed was also maintained to plant in the given and fixed culture medium. The accuracy of the original YOLOv8 was improved to identify the roots, particularly for the diverse root morphology of Phalaenopsis tissue-cultured seedlings. AKConv and DSConv modules were introduced to extract the root features, resulting in an improved AKDS_YOLOv8. Compared with the original YOLOv8, the recognition accuracy of AKDS_YOLOv8 increased by 8.6 percentage point for seedling roots, the recall value increased by 10.7 percentage point, and the average accuracy increased by 7.4 percentage point, when the intersection-over-union ratio of the real box and the detection box was set to 50%. The recognition performance was improved significantly. A dual-input single-output fuzzy was utilized by the platform. The input was taken as the difference between the actual and expected pixel coordinates of the target, as well as the variation in this difference. The membership function was used to obtain the membership degrees of different fuzzy sets, corresponding to the difference and the variation in difference. The resulting data was then fed into the fuzzy control rule table. The image deblurring was employed to determine the acceleration at the end of the robotic arm in the direction of the conveyor belt. 80 repeated planting experiments were conducted to verify the success rate of planting. A success rate of 82.5% was achieved under the failure situations. The platform can effectively pick and plant phalaenopsis tissue-cultured seedlings on the conveyor belt, indicating the promising application value.