No abstract

No abstract

In the automatic apple sorting task, it is necessary to automatically classify certain apple species. A shallow convolutional neural network (CNN) architecture is proposed for this purpose. After collecting a certain number of apple images and labelling them, training data is obtained through a series of data augmentation operations, and then training and parameter optimization are carried out through the Caffe framework. The feasibility of the method is verified by experiments which are divided into two cases. In the case of no occlusion, the classification accuracy of apple images reaches approximately 92% in our test set. Besides, block voting is used to aid the proposed method and a good result can be achieved in our test set in the case of part occlusion caused by branches and leaves, rotten spots, and other kinds of apples. The proposed shallow network is characterized by a small number of parameters and shows resistance to overfitting with a limited dataset. Such a network presents an alternative for classification related tasks in smart visual Internet of Things and brings attention to reducing the complexity of deep neural networks while maintaining their strength. INDEX TERMS Image classification, CNN, overfitting, smart visual Internet of Things.

In order to optimize the pose of welding torch preplanned by offline programming, a structured light-based visual servoing method is proposed. First of all, a series of phase shifting patterns are projected to acquire the so-called phase map. Afterwards, unlike usual feature extraction methods, which were based on 3-D cloud, a cylinder axis is extracted directly from the phase map to represent the connecting pipes' cylindrical surface. Then, a visual servoing control law based on the axis combined with the seam center in phase map is proposed to optimize the pose of the welding torch. Moreover, global asymptotic stability of this method is proved. Finally, simulations and real experiments are performed to demonstrate the effectiveness and robustness of this method. Results show this method can improve the mean error of deviated distance and angle of offline programming by 73.5% and 82.5%, respectively. INDEX TERMSRobotic welding, structured light, feature extraction, robotic positioning, visual servoing, offline programming. JING XU received the Ph.D. degree in mechanical engineering from Tsinghua University, Beijing, China. He was a Postdoctoral Researcher with

Aiming at the problem that the robot de-palletizing task is difficult to accomplish under unstable ambient light, a two-step method is proposed to realize the localization of workpieces, which in this work are woven bags. To begin with, Region Growing method is used to extract the whole target region in the original image, and the relationship model between image intensity and the optimal Region Growing threshold is established. Then, Progressive Probabilistic Hough Transform(PPHT) is used to locate each woven bag. To improve the system performance, the optimal parameters of the PPHT function in different illumination intervals are determined. Finally, experiments are conducted to verify the effectiveness of the proposed method. Experiment results demonstrate this method is robust and feasible.