Search

• A method for apple growth morphology based on Mask RCNN + PointRend has been proposed. • SPPF structure, CA attention mechanism and PANet are introduced as improvements. • CWD algorithm is used for model distillation to improve the detection speed. Identifying apples with different growth forms is an urgent task to be solved by harvesting robots on the path to intelligence. A method based on Mask RCNN + PointRend was proposed for identifying apples with different growth forms in the orchard in the work. The SPPF structure and CA attention mechanism were introduced to the backbone network of the Mask RCNN + PointRend algorithm to improve network performance because of the complexity of the apple orchard. Meanwhile, a bottom-up pathway was introduced into the feature extraction network to enhance feature dissemination, reuse, and fitting. Finally, the number of model parameters was reduced to improve the detection speed, and the CWD algorithm was used for the knowledge distillation of the model. The improved algorithms were called Mask RCNN-Point and Mask RCNN-PointKD. The recognition accuracy of the two enhanced algorithms on the test set was 85.65 and 74.13%, respectively. Besides, average detection time per image was 93.67 and 62.1 ms, respectively. The work provides a reference for identifying apples with different growth forms. [ABSTRACT FROM AUTHOR]

• A visual recognition method of apple growth morphology is proposed. • An improved YOLOv5 deep learning network (called YOLOv5-B) was proposed. • On the basis of YOLOv5, BiFPN-S and ACON-C are introduced as improvements. The work aimed at the visual identification of the growth forms of fruits to facilitate the subsequent use of different harvesting mechanisms for different growth forms of fruits by robots. The improved YOLOv5 deep learning algorithm was used to propose a visual identification method for the growth forms of apples in the orchard. Specifically, the feature extraction module of the YOLOv5 algorithm imitated the BiFPN model to propose the BiFPN-S structure. The spread of features and feature reuse were enhanced to better fit features. The improved algorithm was called YOLOv5-B. The network SiLU activation function was replaced with the ACON-C activation function to improve its network performance. The COCO data set was used to pre-train the network, and then the data set of the work was trained by the transfer learning method. After the training, the generated optimal model was applied for the visual-identification test of the growth of apple fruits. The results showed that the improved algorithm model considered high accuracy and real-time performance, with the map reaching 98.4% and the F1 value of 0.928. The average accuracy of identifying the growth forms of apples for the test set was 98.45%, and the processing speed was 71 FPS. [ABSTRACT FROM AUTHOR]

• Designed a agricultural residual film pollution monitoring system based on UAV. • Implemented autonomous random sampling on agricultural field surfaces using UAV. • The R2 of UAV evaluated film pollution and manual evaluation was 0.966. • The average monitoring and evaluation time of the system is 422.6 s. To evaluate residual film recovery quality and rapidly monitor residual film pollution in agricultural fields, a residual film pollution monitoring system based on UAV, ground station, and cloud server was proposed in this research. It integrated a random sampling point generation (RSPG) algorithm and a key-lock pairing (KLP) algorithm for fast planning of flight paths, and used the U-Net model to rapidly segment residual films from images and calculate residual film coverage. The results showed that the average deviation between the actual distance and the ideal distance of the randomly generated sampling points using the RSPG algorithm was 0.0733 m, and the average deviation of the azimuth angles was 0.2°. The KLP algorithm significantly reduced path planning time, while maintaining a high shortest path rate and reducing memory consumption. The U-Net model achieved an optimal mean intersection over union (MIOU) of 85.63 % on the test set. The simulation and field verification of the system found that automatic UAV path planning and rapid residual film image segmentation were achieved. The coefficient of determination (R2) between the predicted values and the manually labelled values was 0.966, and the root-mean-square error (RMSE) was 0.1173 %. The average relative error between the predicted residual film coverage and the manually labelled coverage was 11.55 %, and the average detection time of the monitoring system was 422.6 s. Therefore, the UAV-based residual film pollution monitoring system can realize integrated automatic sampling and rapid residual film pollution assessment. This study provides technical support for accurate assessment and control of residual film pollution. [ABSTRACT FROM AUTHOR]