We present hierarchical multi-feature classification (HMC) system for multiclass fruit recognition problem. Our approach to HMC exploits the advantages of combining multimodal features and the fruit hierarchy property. In the construction of hybrid features, we take the advantage of using color feature in the fruit recognition problem and combine it with 3D shape feature of depth channel of RGBD (Red, Green, Blue, Depth) images. Meanwhile, given a set of fruit species and variety, with a preexisting hierarchy among them, we consider the problem of assigning images to one of these fruit variety from the point of view of a hierarchy. We report on computational experiment using this approach. We show that the use of hierarchy structure along with hybrid RGBD features can improve the classification performance.
Color histogram has been widely used in feature extraction to represent color feature of an object in the image. In this paper, we identify which features that give high contribution in classification performance, because not all features are directly correlated with object category. In the case of n-bins color histogram, features were referred to color intensity range of color histogram. On the one hand, we consider fruit classification, where the feature space contains various properties of pixel intensities of RGB (Red-Green-Blue) channel. On selecting feature subset, we consider filter method of feature selection. In the filter method, we successively reduce the size of the feature sets and investigate the changes in the classification results. Specifically, we followed the filtering approach to feature selection: selecting features in a single pass first and then applying a classification algorithm independently. We used chi square feature selection to determine relevant features from RGB histogram. Further, we used and evaluated those relevant features in a classification system, using K-Nearest Neighbor (KNN) as classifier. In this paper we show that by conducting feature selection techniques combined with KNN we would be able to prune non-relevant intensities value of Red, Green, and Blue channel. Furthermore, we use the relevant subset of features to identify intensities range of RGB channel that was needed to represent 32 subcategories fruit image efficiently.
11 April 2024 in International Journal of Computational Intelligence Systems
As we all know, cancer is one of the leading causes of death worldwide and the second leading cause of death overall. This is why regular screenings or health checks are necessary to detect cancer lesions early. Since bone scan images have become the primary means of detecting the emergence of cancer lesions on bone, high segmentation accuracy is essential for establishing the model of some predefined regions in bone scan images where cancer metastasis was predicted to appear. Consequently, robust localization and identification of the specific region in bone scan images are required for automated metastasis detection. To this end, we propose Efficient-BtrflyNet, a new deep learning-based architecture for skeleton segmentation of whole-body bone scan images. The proposed architecture exploits the benefits of EfficientNet’s model scaling and the encoder–decoder design of butterfly-type networks. We added EfficientNetB7 to the encoder section to obtain more specific features. The proposed architecture simultaneously processes anterior and posterior whole-body bone scan images. Using 37 bone scan images, we evaluated the performance of our proposed skeleton segmentation system using the Dice score. Efficient-BtrflyNet achieves superior segmentation performance compared to the existing representative method.
15 November 2022 in JIPI (Jurnal Ilmiah Penelitian dan Pembelajaran Informatika)
The use of convolutional neural networks has been applied to various applications. Such as image clas-sification, object detection and recognition, and others. One of the most popular uses for neural networks is image classification. Image classification mainly identifies and categorizes images according to the specified group. One application is to distinguish between one type of dog to another. Classification of dog breeds has its challenges because several kinds of dogs have similar physical characteristics, espe-cially those that belong to the same group. This study explains how to develop a dog breed classification system from a sporting group using a residual neural network (ResNet). The system's goal is to make it simpler for people to identify the dog breed. Five types of dog breeds were used, which were obtained from the Tsinghua Dogs dataset. In its implementation, two variants of CNN are used to be compared, ResNet 50 and ResNet 101, using the same configuration. Based on the research results, ResNet 101 shows better macro-average f1-score results while maintaining high accuracy. The ResNet 50 produces an f1-score of 84%, while ResNet 101 makes an f1-score of 86%.
1 February 2021 in IOP Conference Series: Materials Science and Engineering
Facial and ethnic recognition became a popular research topic. Different from face recognition, ethnic recognition classifies faces according to the general features of certain ethnic groups. Ethnic recognition in face image is increasingly becoming a necessity and is used in various fields. This paper proposed the Indonesian ethnic recognition system based on periorbital features on facial images. We use the five largest ethnic groups in Indonesia in this study, namely Banjar, Bugis, Javanese, Malay, and Sundanese. Gray Level Co-occurrence Matrix and Color Histogram were used as methods, and Random Forest was used as a classifier. Based on - cross-validation tests with optimal k values, the model achieved 98.65% accuracy.
1 March 2023 in IAES International Journal of Artificial Intelligence (IJ-AI)
Autonomous driving is one solution that can minimize and even prevent accidents. In autonomous driving, the vehicle must know the surrounding environment and move under the provisions and situations. We build an autonomous driving system using proximal policy optimization (PPO) in deep reinforcement learning, with PPO acting as an instinct for the agent to choose an action. The instinct will be updated continuously until the agent reaches the destination from the initial point. We use five sensory inputs for the agent to accelerate, turn the steer, hit the brakes, avoid the walls, detect the initial point, and reach the destination point. We evaluated our proposed autonomous driving system in a simulation environment with several branching tracks, reflecting a real-world setting. For our driving simulation purpose in this research, we use the Unity3D engine to construct the dataset (in the form of a road track) and the agent model (in the form of a car). Our experimental results firmly indicate our agent can successfully control a vehicle to navigate to the destination point.