Although one-stage detectors, e.g., YOLOv3, SSD, RefineDet, based on the convolutional neural networks (CNNs) have been widely used in object detection, their localization accuracy of the defects is still poor and the problem of repeated detection still occurs. This paper proposes a surface defects detection method based on improved YOLOv3 for sawn lumbers. Firstly, a Gaussian YOLOv3 is used to estimate the coordinates and the localization uncertainty of the prediction box. Then, the complete intersection over union (CIoU) loss function is adopted to replace the intersection over union (IoU) loss for non-maximum suppression (NMS), because it considers the overlap area, the center point and aspect ratio and thus can reduce the repeated detection. Finally, this paper establishes two datasets, i.e., the rubber lumber dataset containing 4 kinds of defects, and the pine lumber dataset containing 2 kinds of defects. In total, there are 16633 images and 6705 annotated images in these two datasets. The effectiveness of the proposed method has been validated by experiments. Our datasets are available at https://github.com/WenHe-Hnu/sawn-lumber-dataset.

A novel adaptive randomized Hough transform using moving window for circle detection was proposed. In the method, circle detection is done by the standard randomized Hough transform, with a moving window to increase local signal-to-noise ratio for improving the detection rate. Parameters of the randomized Hough transform are set adaptively according to the currently windowed image part to reduce the detecting time while maintaining the user-preferred detection rate. Experimental results on an image database show that the proposed algorithm is effective and superior to other circle detection methods in applications where the range of radii of the circles to be detected is known a priori

There are massive noises and uncertainty factor in traffic data. It is difficult to obtain accurate data from one sensor. Three layers structure to realize the traffic data acquisition based on multi sensors data fusion is given in this article. In the first layer, data from different data source is integrated to an uniform data stream; In the second layer, fuzzy neural network is used to predict the queue length and traffic time. In the third layer, multiphase fuzzy controller for intersection based on queue length is designed. Feasibility of such method is proved by the simulation of one typical intersection

For correcting the color difference in machine vision imaging, all colors in the color space are usually used to train the correction model in traditional methods, which will lead to a complex training process of the model, long training time, and can not achieve high correction precision for each color. In order to solve these problems, a local color space correction scheme is proposed in this paper. In the scheme, the total color space is divided into smaller sub-color space based on color difference threshold and space distance respectively, and the colors in the sub-color space is corrected by polynomial regression method. The experimental results show that local color space correction has higher accuracy and more stable performance than total color space correction.