Megat Syahirul Amin Megat Ali’s research while affiliated with MARA University of Technology and other places

The adoption of precision agriculture in pineapple farming has a significant impact by increasing the yield and reducing the input resources while improving the management of pineapple crops. The intersection of advanced drone technology and cutting-edge artificial intelligence has reformed fruit crop management through revolutionary levels of automation, precision fruit detection, yield estimation, and crop health detection. However, the capability for obscuring the detection of subtle features to better manage occlusions and complex environments in images captured by drones at certain heights with drones is challenging to distinguish, thus hindering an accurate object analysis for fruit-environment differentiation. The proposed work uses Deep Learning (DL) techniques to classify pineapple fruit images captured ten meters above the ground. This is achieved specifically through the use of pretrained models and Faster Region-Based Convolutional Neural Networks (Faster R-CNNs) due to their ability to learn robust interpretations from images for object classification tasks. This paper evaluates the capabilities and accuracies of four pretrained models, namely ResNet-101, ResNet-50, Inception-ResNet-v2, and VGG-19, to detect and classify the pineapple fruit amidst the complex background and varying lighting conditions. By evaluating the pretrained models for pineapple fruit classification using comprehensive metrics (True Positive Rate (TPR), False Positive Rate (FPR), Accuracy (ACC), Recall (REC), Precision (PRE), F1-score), the results reveal that the Faster R-CNN architecture with the VGG-19 pretrained model outperformed the other architectures, demonstrating the best performance in pineapple fruit detection with an ACC of 0.7924 (79.24%), a PRE of 0.9990 (99.90%), a REC of 0.7930 (79.30%), and an F1-score of 0.8839 (88.39%). The effectiveness of this model in overseeing complex scenarios suggests potential improvements in classification accuracy compared to other pretrained models, while acknowledging performance variability across various architectures.

It is vital to combat crimes by predicting and detecting the occurrence of crime, especially in urban cities. Hence this study proposed investigating the capability of six deep learning models, namely the AlexNet, GoogleNet, ResNet18, ResNet50, ResNet101 and InceptionV3, in determining the most optimum model for snatch theft detection. Two categories of databases comprising 13000 images of snatch theft and non-snatch activities were generated from 120 videos obtained from the Google and YouTube platforms. These images are further used for training and testing these six DL models, along with data augmentation implemented during training to avoid overfitting. However, it was found that overfitting occurred based on training and testing accuracy plots, and hence, it was decided to re-train the model using an early stopping method. Thus, upon completion of re-training all six models, it was found that all six models showed a good-fit condition, with ResNet 50 attaining the highest testing accuracy of 98.9% and 100% sensitivity. As for specificity, ResNet 101 showed the highest value, precisely 97.7%.

This study has proposed a non-linear autoregressive model to predict one-day ahead dissolved oxygen in paddy field irrigation channel. A 32-day data is obtained from Kampung Padang To’ La in Pasir Mas, Kelantan using off-the shelf water quality parameter sensors. Analysis has revealed no correlation between dissolved oxygen with pH and electrical conductivity. A non-linear autoregressive model is then developed using the dissolved oxygen measurements and artificial neural network. A prediction model developed using Levenberg- Marquardt algorithm yielded the best results with overall regression of 0.9253. The model has also passed all correlation tests and can therefore, be accepted.

Osteogenesis Imperfecta (OI) image segmentation by using Deep Convolutional Neural Network (DCNN) is yet to be evaluated. The segmentation of OI is very important as a useful tool for medical experts to further analyze the fracture risk and avoid bone fractures. In this paper, we present the review of DCNN architecture used in image segmentation. The images were obtained from different types of modalities such as Computed Tomography (CT), Magnetic Resonance Imaging (MRI), or Ultrasound. Several architectures have been used by previous studies include U-Net, faster R-CNN, ResNet, and MS-Net architecture to automatically segment the images. Overall, all researchers from the reviewed papers concluded that the proposed DCNN architecture gave good performance results.KeywordsDeep learningDCNN architectureImage segmentation

p>Robotic prosthetics is increasingly adopted as an enabling technology for amputees. These are vital not only for activities of daily living but to display expression and affection. A vital element to this system is an intelligent model that can identify signatures from the remaining limb that can be mapped to specific effector movements. Therefore, the study proposes the use of forearm electromyogram to classify between different types of hand gestures; fingers spread, wave out, wave in, fist, double tap, and relaxed state. Data are acquired from 32 subjects using Myo armband. Initially, a total of 248 time-and frequency-domain features are extracted from the eightchannel device. Neighborhood component analysis has reduced them to a total of fourteen features. A hand gesture classification model based on electromyogram signal has been successfully developed using support vector machine with overall accuracy of 97.4% for training, and 88.0% for testing.</p

Osteogenesis Imperfecta (OI) is a bone disorder that causes bone to be brittle and easy to fracture. The patient suffered from this disease will have poor quality of life. Simulation on the bone fracture risk would help medical doctors to make decision in their diagnosis. Detection of edges from the OI images is very important as it helps radiologist to segmentize cortical and cancellous bone to make a good 3D bone model for analysis. The purpose of this paper is to review the fundamentals of fuzzy logic in edge detection of OI bone as it is yet to be implemented. Several fuzzy logic concepts are reviewed by previous studies which include fuzziness, membership functions and fuzzy sets regarding digital images. The OI images were produced by modalities such as Magnetic Resonance Imaging (MRI), Ultrasound, or Computed Tomography (CT). In summary, researchers from the reviewed papers concluded that fuzzy logic can be implemented to detect edges in noisy clinical images.