SSD (Single Shot Multibox Detector) is one of the best object detection algorithms with both high accuracy and fast speed. However, SSD's feature pyramid detection method makes it hard to fuse the features from different scales. In this paper, we proposed FSSD (Feature Fusion Single Shot Multibox Detector), an enhanced SSD with a novel and lightweight feature fusion module which can improve the performance significantly over SSD with just a little speed drop. In the feature fusion module, features from different layers with different scales are concatenated together, followed by some down-sampling blocks to generate new feature pyramid, which will be fed to multibox detectors to predict the final detection results. On the Pascal VOC 2007 test, our network can achieve 82.7 mAP (mean average precision) at the speed of 65.8 FPS (frame per second) with the input size 300×300 using a single Nvidia 1080Ti GPU. In addition, our result on COCO is also better than the conventional SSD with a large margin. Our FSSD outperforms a lot of state-of-the-art object detection algorithms in both aspects of accuracy and speed. Code is available at https://github.com/ lzx1413/CAFFE_SSD/tree/fssd.

The automatic lung nodule detection system can facilitate the early screening of lung cancer and timely medical interventions. However, there still exist multiple nodule candidates produced by initial rough detection in this system, and how to determine authenticity is a key problem. As this work is often challenged by the radiological heterogeneity of the computed tomography scans and the variable sizes of lung nodules, we put forward a multi-resolution convolutional neural network (CNN) to extract features of various levels and resolutions from different depth layers in the network for classification of lung nodule candidates. Through the use of knowledge transfer, the method can be divided into three steps. First, we transfer knowledge from the source CNN model which has been applied to edge detection and improve the model to a new multi-resolution model which is suitable for the image classification task. Then, the knowledge is transformed from source training progress so that all of the side-output branches in the model will be considered in the calculation. Moreover, the loss function and objective equation are improved to be imagewise calculation rather than pixel-wise. Finally, samples production and data enhancement are performed to train and test a classifier tailored for classification of lung nodule candidates. The experimental results on the LUNA16 data set show that our method gets an accuracy of 0.9733, a precision of 0.9673, and an AUC of 0.9954 while being used for lung nodule candidate classification, which is higher than the scores obtained by most of the state-of-the-art approach. In addition, when the test samples with three different sizes of 26 * 26, 36 * 36, and 48 * 48 are used to test the multi-resolution CNN, the accuracy rate of all three experiments exceed 92.81%, which demonstrates that the proposed model is insensitive to input scales. INDEX TERMS Convolutional neural network, lung nodule candidate classification, multi-resolution model, knowledge transfer.

No abstract

No abstract

Freight car target detection plays an important role in railway traffic safety, which typically depends on artificial observation or conventional machine learning, with insufficient accuracy and high demand for an observer's physical strength and image quality. Motivated by the recent advances of the convolutional neural network in object detection, this study investigates how deep neural networks can be applied in freight car target detection to better solve the aforementioned problems. We propose a novel two-training method for freight car target detection; the method includes general training and special training. In addition, online hard example mining and deformable convolutional network are introduced to select hard examples and extract better features for the special training stage to improve the problem of tiny target detection in poor images obtained from freight car target detection. The proposed methods are verified using experimental results based on three aspects, i.e. indexes, visualization, and speed. High accuracy can be achieved with good recall and acceptable speed for freight car target detection applications. Finally, we illustrate the utility of using such a model to test high robustness for changes in image quality and other target detection tasks with slight modification.