Browsing by Author "Ferdowsy, Faria"

Now showing 1 - 3 of 3

A Computer Vision Based Food Recognition Approach for Controlling Inflammation to Enhance Quality of Life of Psoriasis Patients
(2021 International Conference on Information Technology (ICIT), IEEE, 2021-07-26) Hridoy, Rashidul Hasan; Akter, Fatema; Mahfuzullah, Md.; Ferdowsy, Faria
Deep learning becomes the spotlight in computer vision based recognition approaches in recent years. Psoriasis affects people of all ages around the world and causes inflammation on the skin with significant systemic disability and illness. Inflammatory foods increase inflammation rapidly, patients can easily control inflammation to enhance the quality of life by eliminating these foods from their everyday diet. This paper addresses a rapid food recognition approach to assist psoriasis patients to recognize fifteen highly inflammatory foods. Using image augmentation techniques, a dataset of 41250 images of different inflammatory foods have generated from 10000 images. AlexNet, VGG16, and EfficientNet-B0 have used in this study using the transfer learning approach, and EfficientNet-B0 has achieved the highest accuracy of 98.63% under the test set of 5250 images. AlexNet and VGG16 have achieved 87.22% and 93.79% accuracy, respectively. EfficientNet-B0 has consumed the lowest time in recognizing unseen images compared to others.
An Early Recognition Approach for Okra Plant Diseases and Pests Classification Based on Deep Convolutional Neural Networks
(Scopus, 2021) Hridoy, Rashidul Hasan; Afroz, Maisha; Ferdowsy, Faria
The issue of effective plant disease and pest prevention is compactly connected to the issues of sustainable agronomics and climate change. Okra plant diseases and pests cause intense monetary losses to the growing okra industry, but their accurate and rapid identification remains troublesome due to the lack of efficient approaches. This paper addresses an early recognition approach for controlling the disease and pest spread to ensure quality production of okra. At first, a dataset of fifteen classes is generated from 12476 collected images using nine image augmentation techniques which contains 124760 images of okra plant diseases and pests. Afterwards, state-of-the-art deep learning models such as InceptionResNetV2, Xception, ResNet50, MobileNetV2, VGG16, and AlexNet were utilized with the transfer learning approach. InceptionResNetV2 showed significant performance compared to others, achieved 98.73% and 98.16% accuracy under the training set of 99808 images, and the test set of 6236 images of the used dataset, respectively.
Obesity Risk Prediction Using Machine Learning
(Daffodil International University, 2021-01-31) Ferdowsy, Faria; Fatema, Kaniz; Yeasmin, Tammim
Obesity is dangerous for health. Nowadays it has become a threat to people all over the world. The range of obese people in Bangladesh is also increasing rapidly and especially young people are being attacked because of their addiction to junk food. Obesity has a negative impact on our body and life. We have to keep an eye on the topic of people being obese and their health and body getting affected by it which can make them have many deadly diseases and drag to death also. We need to stay away from an unhealthy lifestyle, should stop overeating, should move our body enough, and try to avoid being obese. We will predict the risk of becoming obese with machine learning. First, we study some related papers, journals, and online articles then we talk to doctors and obese people; we find some common factors that are related to become obese. Then we collect data based on those factors, such as genetics, age, height, weight, diet, gender, profession, health ability, mental pressure, trauma, daily life routine, etc. We collect data from both types of people who are obese and who are not obese. We have two outcomes and the outcomes are ‘yes’, and ‘no’. These will describe that if there is any extra fat is accumulating on the body or not. After data collection, we processed all the data and created a processed dataset. We applied machine-learning algorithms to our processed dataset. Since machine learning, artificial intelligence, and deep learning used in various predictions and detection systems. We used k-nearest neighbor, logistic regression, support vector machine (SVM), naïve Bayes, random forest, adaptive boosting (ADA boosting), decision tree, multilayer perceptron (MLP), and gradient boosting classifier. In our work, out of the nine algorithms, logistics regression gave the best performance based on accuracy and the accuracy of logistic regression was 97.03%.