SurfCNN: A Descriptor Accelerated Convolutional Neural Network for Image-Based Indoor Localization

Abstract

Convolutional neural network (CNN) is a powerful tool for many data applications. However, its high dimension nature, large network size and computational complexity, and the need of large amount of training data make it challenging to be used in edge computing applications, which are becoming increasingly popular, relevant and important. In this paper, we propose a descriptor based approach to accelerate convolutional neural network training, reduce input dimension and network size, which greatly facilitates the use of CNN for edge computating and even cloud computing. By using image descriptors to extract features from original images, we report a simpler convolutional neural network with fast training speed, low memory usage and outstanding accuracy without the need for a pre-trained network as opposed to the state of art. In indoor localization, our SURF descriptors accelerated CNN (SurfCNN) can reach an average position error of 0.28 m and orientation error of 9.2 •. Compared to the conventional CNN that uses original images as input, our algorithm reduces the dimension of the input features by a factor of 48 without impairing the accuracy. Further, at an extreme feature reduction of 14,440 times, our model still retains an average position error retained at 0.41 m and orientation error at 14 • .

A CNN-LSTM network with multi-level feature extraction-based approach for automated detection of coronavirus from CT scan and X-ray images

“…It needs a high-dimensional optimization method during which the training time is long since the input size is large. In addition, if testing samples are drastically different from training samples, then the CNN network must be retrained or fine-tuned for identification [27] , [28] . Level 1 feature extraction is used to reduce input dimensions of data for the CNN network.…”

“…However, in the proposed approach, we consider the matrix formed by the handcrafted feature vectors instead of a direct image as an input of the CNN network. Image pixels usually connect locally and globally; similarly, the fused feature vectors represent the locally and globally correlated features [27] , [28] . Therefore, we choose a fused feature set as an input of the CNN network.…”

“…However, the training process of these networks is more complex, time-consuming, and resource-consuming due to high dimensional input. As a result, the need for these networks finishes, especially in IoT-based medical solutions where the computing resources are minimal compared to CNN computational complexity [27] , [28] .…”

Untitled

“…Implementations of real-time feature extraction can be achieved using different hardware platforms such as : Graphics Processing Units (GPUs) , multi-core processors , Application Specific Integrated Circuits (ASICs) , and FPGAs [6]. The multi-core processors and GPUs are commonly used to speed up the computation of image processing and computer vision algorithms [7][9] [12]. ASICs and FPGAs are flexible and more power efficient.…”

Android Device Malware Classification Framework Using Multistep Image Feature Extraction and Multihead Deep Neural Ensemble

“…However, the training process of these networks is complex, time consuming, and resource consuming because of high dimensional input. As a result, the need for these networks must be removed, especially in Internet-of-Things (IoT)-based solutions where the computing resources are less compared with CNN computational complexity [10,11]. Second, there is no guarantee that the classifier can generalize to new or unknown data.…”