Deep Rule-Based Aerial Scene Classifier using High-Level Ensemble Feature Descriptor

Gu, Xiaowei; Angelov, Plamen

doi:10.1109/ijcnn.2019.8851838

Cited by 4 publications

(3 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…4, where each input image is segmented to five sub-images (centre and four corners) and the VGG-VD-16 model is used for feature extraction. The architecture used in [76] is involved in the experiment as Arch. By comparing between Arch.…”

Section: Numerical Resultsmentioning

confidence: 99%

Highly interpretable hierarchical deep rule-based classifier

Angelov

2020

Applied Soft Computing

Self Cite

View full text Add to dashboard Cite

Deep learning paradigm is arguably the fastest growing branch of machine learning and artificial intelligence (AI) at the moment [1,2]. Deep neural networks (DNNs) are entirely based on the artificial neural networks and probabilistic type of uncertainties [3]. They have demonstrated eye-catching successes on image classification [4,5], speech processing [6,7] and many other complex problems [8,9] that traditional machine learning approaches are struggling with.Despite of the impressive advances DNNs have achieved, the research communities and enterprises are increasingly demanding explainable AI [10,11]. Indeed, DNN-based AI algorithms nowadays are more frequently getting involved for making decisions in financial and safety-critical applications [12,13]. However, DNNs are the typical type of "black box" models with a very high level of complexity level that only computers can understand. It is reported that such "black box" models can provide a wrong outcome with high confidence by modifying just one pixel in the input images [14]. The lack of transparency and explainability can pose a significant obstacle, especially for highly regulated, high-risk/high-value industries. Therefore, there is a high demand in developing alternative architectures, learning and model structure paradigms that can provide offer [15]: (1) high levels of precision comparable or surpassing the level achieved by humans or/and by the state-of-the-art methods (including DNNs); (2) be highly transparent, interpretable, easy to explain and use for humans; (3) computationally efficient, fast to train and use; (4) computational resource and training data lean -able to be trained with a single or handful or examples per class, not requiring computer accelerators, such as GPU, HPC, etc.As one of the pillars of the computational intelligence, fuzzy rule-based (FRB) systems are a mathematical tool to describe the human reasoning and decision-making processes [11]. FRB systems take the form of zero-order, Highly interpretable hierarchical deep rule-based classifierXiaowei Gu a, b, ⁎, 1

show abstract

Section: Numerical Resultsmentioning

confidence: 99%

Highly interpretable hierarchical deep rule-based classifier

Angelov

2020

Applied Soft Computing

Self Cite

View full text Add to dashboard Cite

show abstract

“…The second module of SeRBIA is an ensemble of pretrained DCNN models for feature extraction [34]. In accordance with our previous research [35], both the AlexNet [32] and VGG-VD-16 [33] models are employed to create an ensemble feature descriptor. This is because this particular combination has demonstrated stronger descriptive abilities and results in a higher accuracy in classifying aerial images compared with other benchmark approaches.…”

Section: General Architecturementioning

confidence: 90%

A Semi-Supervised Deep Rule-Based Approach for Complex Satellite Sensor Image Analysis

Angelov

Zhang

et al. 2021

IEEE Trans. Pattern Anal. Mach. Intell.

Self Cite

View full text Add to dashboard Cite

Large-scale (large-area), fine spatial resolution satellite sensor images are valuable data sources for Earth observation while not yet fully exploited by research communities for practical applications. Often, such images exhibit highly complex geometrical structures and spatial patterns, and distinctive characteristics of multiple land-use categories may appear at the same region. Autonomous information extraction from these images is essential in the field of pattern recognition within remote sensing, but this task is extremely challenging due to the spectral and spatial complexity captured in satellite sensor imagery. In this research, a semi-supervised deep rule-based approach for satellite sensor image analysis (SeRBIA) is proposed, where large-scale satellite sensor images are analysed autonomously and classified into detailed land-use categories. Using an ensemble feature descriptor derived from pre-trained AlexNet and VGG-VD-16 models, SeRBIA is capable of learning continuously from both labelled and unlabelled images through self-adaptation without human involvement or intervention. Extensive numerical experiments were conducted on both benchmark datasets and real-world satellite sensor images to comprehensively test the validity and effectiveness of the proposed method. The novel information mining technique developed here can be applied to analyse large-scale satellite sensor images with high accuracy and interpretability, across a wide range of real-world applications.

show abstract

“…Gu and Angelov [24] have proposed a new deep rule-based approach using a high-level ensemble feature descriptor for aerial scene classi cation. They have ensembled three pre-trained deep convolutional neural networks for feature extraction and extracted more discriminative representations from the local regions of aerial images.…”

Section: Related Workmentioning

confidence: 99%