On the Behavior of Convolutional Nets for Feature Extraction

Abstract

Deep neural networks are representation learning techniques. During training, a deep net is capable of generating a descriptive language of unprecedented size and detail in machine learning. Extracting the descriptive language coded within a trained CNN model (in the case of image data), and reusing it for other purposes is a field of interest, as it provides access to the visual descriptors previously learnt by the CNN after processing millions of images, without requiring an expensive training phase. Contributions to this field (commonly known as feature representation transfer or transfer learning) have been purely empirical so far, extracting all CNN features from a single layer close to the output and testing their performance by feeding them to a classifier. This approach has provided consistent results, although its relevance is limited to classification tasks. In a completely different approach, in this paper we statistically measure the discriminative power of every single feature found within a deep CNN, when used for characterizing every class of 11 datasets. We seek to provide new insights into the behavior of CNN features, particularly the ones from convolutional layers, as this can be relevant for their application to knowledge representation and reasoning. Our results confirm that low and middle level features may behave differently to high level features, but only under certain conditions. We find that all CNN features can be used for knowledge representation purposes both by their presence or by their absence, doubling the information a single CNN feature may provide. We also study how much noise these features may include, and propose a thresholding approach to discard most of it. All these insights have a direct application to the generation of CNN embedding spaces.

Automated Systems for Detection of COVID-19 Using Chest X-ray Images and Lightweight Convolutional Neural Networks

“…In this paper, the features are extracted from the FC layer of three architectures (AOCT-Net, MobileNet, and ShuffleNet) which is a common method since the FC layer proceeds the Softmax classifier. Based on the selected layer only three features from each class will be extracted and these features will a fine selected and representative [49,50]. The scatter distribution for the extracted features using three models is represented in Figure 5.…”

“…In this way, the learning rule in a Softmax classifier for the binary units is similar to the regular binary unit law. The only difference is that the Softmax function model is the generalization of the logistic sigmoid function, which can handle classification problems with more than two possible values [49,50].…”

Studying the impact of the Full-Network embedding on multimodal pipelines

“…The mapping of standardized values into these three categories is done through the definition of two constant thresholds. The optimal values of these thresholds can be found empirically for a labeled dataset [33]. However, we use certain threshold values shown to perform consistently across several domains [4].…”

Land Cover Segmentation of Airborne LiDAR Data Using Stochastic Atrous Network

“…Deep learning methodologies contribute to simplifying the feature engineering process, and by using a deep multi-layer convolutional neural network (CNN) [16], the features are learned from the data during the training process. Consequently, a successfully trained CNN model can be considered as a feature extractor that both combines features in different spatial locations and takes into account the aspects of spatial autocorrelation between features.…”