Convolutional neural networks have recently experienced successful development in the field of computer vision. In precision agriculture, apple picking robots use computer vision methods to detect apples in orchards. However, existing object detection algorithms often face problems such as leaf shading, complex illumination environments, and small, dense recognition targets, resulting in low apple detection rates and inaccurate localization. In view of these problems, we designed an apple detection model based on lightweight YOLOv4—called Improved YOLOv4—from the perspective of industrial application. First, to improve the detection accuracy while reducing the amount of computation, the GhostNet feature extraction network with a Coordinate Attention module is implemented in YOLOv4, and depth-wise separable convolution is introduced to reconstruct the neck and YOLO head structures. Then, a Coordinate Attention module is added to the feature pyramid network (FPN) structure in order to enhance the feature extraction ability for medium and small targets. In the last 15% of epochs in training, the mosaic data augmentation strategy is turned off in order to further improve the detection performance. Finally, a long-range target screening strategy is proposed for standardized dense planting apple orchards with dwarf rootstock, removing apples in non-target rows and improving detection performance and recognition speed. On the constructed apple data set, compared with YOLOv4, the mAP of Improved YOLOv4 was increased by 3.45% (to 95.72%). The weight size of Improved YOLOv4 is only 37.9 MB, 15.53% of that of YOLOv4, and the detection speed is improved by 5.7 FPS. Two detection methods of similar size—YOLOX-s and EfficientNetB0-YOLOv3—were compared with Improved YOLOv4. Improved YOLOv4 outperformed these two algorithms by 1.82% and 2.33% mAP, respectively, on the total test set and performed optimally under all illumination conditions. The presented results indicate that Improved YOLOv4 has excellent detection accuracy and good robustness, and the proposed long-range target screening strategy has an important reference value for solving the problem of accurate and rapid identification of various fruits in standard orchards.

Branch identification is key to the robotic pruning system for apple trees. High identification accuracy and the positioning of junction points between branch and trunk are important prerequisites for pruning with a robotic arm. Recently, with the development of deep learning, Transformer has been gradually applied to the field of computer vision and achieved good results. However, the effect of branch identification based on Transformer has not been verified so far. Taking Swin-T and Resnet50 as a backbone, this study detected and segmented the trunk, primary branch and support of apple trees on the basis of Mask R-CNN and Cascade Mask R-CNN. The results show that, when Intersection over Union (IoU) is 0.5, the bbox mAP and segm mAP of Cascade Mask R-CNN Swin-T are the highest, which are 0.943 and 0.940; as for the each category identification, Cascade Mask R-CNN Swin-T shows no significant difference with the other three algorithms in trunk and primary branch; when the identified object is a support, the bbox AP and segm AP of Cascade Mask R-CNN Swin-T is significantly higher than that of other algorithms, which are 0.879 and 0.893. Next, Cascade Mask R-CNN SW-T is combined with Zhang & Suen to obtain the junction point. Compared with the direct application of Zhang & Suen algorithm, the skeleton obtained by this method is advantaged by trunk diameter information, and its shape and junction points position are closer to the actual apple trees. This model and method can be applied to follow-up research and offer a new solution to the robotic pruning system for apple trees.

Spraying is the most widely used means of pesticide application for pest control in agriculture and forestry. The atomization characteristics of the nozzles are directly related to the spray drift, rebound, and deposition. Previous research studies have mainly focused on the change pattern of atomization characteristics. Mathematical descriptions of the atomization characteristics of flat fan nozzles are rare, and pesticide application theories are also insufficient. Atomization characteristics mainly include droplet size and velocity. This study analyzes the influence of the spray parameters (spray angle, pressure, and equivalent orifice diameter of nozzles) and the spatial position in the flow field. To obtain the atomization characteristics of flat fan nozzles, the phase Doppler particle analyzer (PDPA) was selected for the accurate measurement of the droplet sizes and velocities at distances 0.30–0.60 m, using low spray pressures (0.15–0.35 MPa). The droplet size and velocity models were then established and validated. The results revealed that the average absolute error of the droplet size model was 23.74 µm and the average relative error was 8.23%. The average absolute and relative errors of the droplet velocity model were 0.37 m/s and 7.86%, respectively. At a constant spray pressure and angle, there was a positive correlation between the droplet size and the equivalent orifice diameter of the nozzles. The test also verified that the spray angle and distance had a negative correlation with the droplet velocity at a given pressure. The spray distance had no effect on the spray axial droplet size at constant spray pressure. In addition, the spray angle greatly affected the droplet velocity along the X-axis; similarly, the spray parameters, especially spray angle, greatly affected the droplet size.

The characteristic parameters of Citrus grandis var. Longanyou canopies are important when measuring yield and spraying pesticides. However, the feasibility of the canopy reconstruction method based on point clouds has not been confirmed with these canopies. Therefore, LiDAR point cloud data for C. grandis var. Longanyou were obtained to facilitate the management of groves of this species. Then, a cloth simulation filter and European clustering algorithm were used to realize individual canopy extraction. After calculating canopy height and width, canopy reconstruction and volume calculation were realized using six approaches: by a manual method and using five algorithms based on point clouds (convex hull, CH; convex hull by slices; voxel-based, VB; alpha-shape, AS; alpha-shape by slices, ASBS). ASBS is an innovative algorithm that combines AS with slices optimization, and can best approximate the actual canopy shape. Moreover, the CH algorithm had the shortest run time, and the R² values of V_CH, V_VB, V_AS, and V_ASBS algorithms were above 0.87. The volume with the highest accuracy was obtained from the ASBS algorithm, and the CH algorithm had the shortest computation time. In addition, a theoretical but preliminarily system suitable for the calculation of the canopy volume of C. grandis var. Longanyou was developed, which provides a theoretical reference for the efficient and accurate realization of future functional modules such as accurate plant protection, orchard obstacle avoidance, and biomass estimation.

The tea industry is one of China’s most important industries. The picking of famous tea still relies on manual methods, with low efficiency, labor shortages and high labor costs, which restrict the development of the tea industry. These labor-intensive picking methods urgently need to be transformed into intelligent and automated picking. In response to difficulties in identification of tea buds and positioning of picking points, this study took the one bud with one leaf grade of the Fuyun 6 tea species under complex background as the research object, and proposed a method based on deep learning, combining object detection and semantic segmentation networks, to first detect the tea buds, then segment the picking area from the tea bud detection box, and then obtain the picking point from the picking area. An improved YOLOX-tiny model and an improved PSP-net model were used to detect tea buds and their picking areas, respectively; the two models were combined at the inference end, and the centroid of the picking area was taken as the picking point. The YOLOX-tiny model for tea bud detection was modified by replacing its activation function with the Mish function and using a content-aware reassembly of feature module to implement the upsampling operation. The detection effects of the YOLOX-tiny model were improved, and the mean average precision and recall rate of the improved model reached 97.42% and 95.09%, respectively. This study also proposed an improved PSP-net semantic segmentation model for segmenting the picking area inside a detection box. The PSP-net was modified by replacing its backbone network with the lightweight network MobileNetV2 and by replacing conventional convolution in its feature fusion part with Omni-Dimensional Dynamic Convolution. The model’s lightweight characteristics were significantly improved and its segmentation accuracy for the picking area was also improved. The mean intersection over union and mean pixel accuracy of the improved PSP-net model are 88.83% and 92.96%, respectively, while its computation and parameter amounts are reduced by 95.71% and 96.10%, respectively, compared to the original PSP-net. The method proposed in this study achieves a mean intersection over union and mean pixel accuracy of 83.27% and 86.51% for the overall picking area segmentation, respectively, and the detecting rate of picking point identification reaches 95.6%. Moreover, its detection speed satisfies the requirements of real-time detection, providing a theoretical basis for the automated picking of famous tea.