Modified Capsule Neural Network (Mod-CapsNet) for Indoor Home Scene Recognition

Basu, Amlan; Kaewrak, Keerati; Petropoulakis, Lykourgos; Caterina, Gaetano Di; Soraghan, John J.

doi:10.1109/ijcnn48605.2020.9207084

Cited by 6 publications

(3 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In this work, a CapsNet with a pooling layer is presented for improved accuracy -this approach was also shown to provide superior results in our study [2]. Nonetheless, any 3D neural network will, inevitably, have a large number of parameters requiring training.…”

Section: Introductionmentioning

confidence: 72%

“…To mitigate this issue the development of 1D CapsNetA and 1D CapsNetB is proposed in this paper. Further, the paper concentrates more on making CapsNets better because they can produce good accuracy on smaller datasets [1,2]. Therefore, the aforementioned works (apart from CapsNets) discussed are to establish a proper comparison with the work presented in this paper.…”

Section: Background Workmentioning

confidence: 97%

See 1 more Smart Citation

3-Dimensional Object Recognition Using 1-Dimensional Capsule Neural Networks

Basu

Kaewrak

Petropoulakis

et al. 2022

2022 International Conference on Emerging Techniques in Computational Intelligence (ICETCI)

View full text Add to dashboard Cite

Currently, indoor home object recognition systems lack the degree of accuracy required for reliable automated operations. In this paper, a 3-Dimensional (3D) object recognition deep neural network system, capable of recognizing indoor objects from 3D images with a view to assisting indoor robotic devices in performing home tasks, is presented. Almost invariably such systems drastically increase the number of total parameters which must be trained leading to very timeconsuming training processes. Furthermore, the lack of suitably large annotated datasets for indoor objects adds to the training difficulties. To address these challenges a 3D dataset arranged in a 1D array format along with new architectures of 1D CapsNet is proposed. This effectively reduces the total number of parameters, resulting in faster and more accurate training of a neural network.Using the proposed architecture also appears to improve the training accuracy even when the datasets are relatively small. For this work, ModelNet-10 and the ModelNet-40 datasets are used. They have indoor home object images in 3D, which are few when compared to other datasets.

show abstract

Section: Introductionmentioning

confidence: 72%

Section: Background Workmentioning

confidence: 97%

3-Dimensional Object Recognition Using 1-Dimensional Capsule Neural Networks

Basu

Kaewrak

Petropoulakis

et al. 2022

2022 International Conference on Emerging Techniques in Computational Intelligence (ICETCI)

View full text Add to dashboard Cite

show abstract

“…Future research, for instance, has been utilizing a linear combination approach between capsules, which reduces the number of capsules while enhancing their capacity to represent seen things. Another study uses a modified capsule neural network with a pooling layer and no squash function to identify [8] grayscale inside home scenes. As compared to a traditional Caps Net's accuracy of 17.2%, our Mod-Caps Net delivered results with a 70% accuracy [9] to circumvent the challenge of classifying the task of verifying the kind of font used in a file.…”

Section: Related Workmentioning

confidence: 99%

Survey On Medical Image Classification Using CAPSGNN

Shanmugam¹,

P²,

Ma³

2023

RRRJ

View full text Add to dashboard Cite

The general Convolutional Neural Networks (CNNs) have been in practice, being the most conventional algorithm for image-based detection and classification. But over the years, after extensive use of CNN algorithms with different architectures, it has been shown that CNN tends to lose details and features of the image. This led to the use of Capsule-based neural networks for image detection and classification. On the other side, CNN has evolved and integrated with another type of neural network called the Graph Neural Network (GNN). Many existing systems have drawbacks such as feature loss and computation efficiency. Several transfer learning models have been introduced to solve these problems by modifying the existing models and adding different combinations of layers and hyper parameters. However, they still don't provide a clear solution as they are just derived algorithms. Therefore, there is a need to design an algorithm and technique that approaches the image classification process in a unique and different way. This is where the CAPSGNN algorithm comes into use. This proposed model uses the best features of all the other algorithms and fuses them into one algorithm. This reduces the computation time and solves the feature loss problems. Now, reports can be generated faster and more accurately for assisting the process of disease diagnosis in hospitals and saving doctors' time spent on reviewing every report. These speeds up the cycle of the medical field, as the identification of diseases takes more time than the actual treatment and needs to be processed faster for faster treatment and recovery.

show abstract