Feature Segmentation Research Articles

Predicting Paediatric Brain Disorders from MRI Images Using Advanced Deep Learning Techniques.

The problem at hand is the significant global health challenge posed by children's diseases, where timely and accurate diagnosis is crucial for effective treatment and management. Conventional diagnosis techniques are typical, use tedious processes and generate inaccurate results since they are executed by human beings and cause delays in treatment that can be fatal. Considering these and other shortcomings there exists a need to have more efficient and accurate solutions based on artificial intelligence. Machine learning and more specifically, deep learning algorithms are of great help in analysing medical and clinical images to detect as well as classify diseases. In this paper, we propose a system for detecting various childhood diseases using a range of advanced Convolutional Neural Network models like EfficientNetB0, EfficientNetB3, Xception, InceptionV3, MobileNetV2, VGG19, DenseNet169, ResNet50V2, ResNet152V2, and the hybrid architecture InceptionResNetV2. These models are trained on MRI images of paediatric brain disorders to achieve high prediction accuracy. We use data visualization techniques such as segmentation and contour-based feature extraction to extract regions of interest before feeding the data into the models. The models are optimized using both ADAM and RMSprop optimizers. EfficientNetB0, when optimized with RMSprop, achieves an accuracy of 94.59%, a loss of 0.44, and an RMSE of 0.66. InceptionResNetV2, optimized with ADAM, achieves the highest accuracy of 97.59%, while EfficientNetB0 demonstrates the lowest loss (0.25) and RMSE (0.5). We also evaluate the models based on their precision, learning curves, recall, computational time, and F1 score, highlighting the effectiveness of AI-driven approaches for the diagnosis and management of children's diseases.

Bird Species Detection Net: Bird Species Detection Based on the Extraction of Local Details and Global Information Using a Dual-Feature Mixer.

Bird species detection is critical for applications such as the analysis of bird population dynamics and species diversity. However, this task remains challenging due to local structural similarities and class imbalances among bird species. Currently, most deep learning algorithms focus on designing local feature extraction modules while ignoring the importance of global information. However, this global information is essential for accurate bird species detection. To address this limitation, we propose BSD-Net, a bird species detection network. BSD-Net efficiently learns local and global information in pixels to accurately detect bird species. BSD-Net consists of two main components: a dual-branch feature mixer (DBFM) and a prediction balancing module (PBM). The dual-branch feature mixer extracts features from dichotomous feature segments using global attention and deep convolution, expanding the network's receptive field and achieving a strong inductive bias, allowing the network to distinguish between similar local details. The prediction balance module balances the difference in feature space based on the pixel values of each category, thereby resolving category imbalances and improving the network's detection accuracy. The experimental results using two public benchmarks and a self-constructed Poyang Lake Bird dataset demonstrate that BSD-Net outperforms existing methods, achieving 45.71% and 80.00% mAP50 with the CUB-200-2011 and Poyang Lake Bird datasets, respectively, and 66.03% AP with FBD-SV-2024, allowing for more accurate location and species information for bird detection tasks in video surveillance.

Cross-reactive sarbecovirus antibodies induced by mosaic RBD-nanoparticles.

Therapeutic monoclonal antibodies (mAbs) against SARS-CoV-2 become obsolete as spike substitutions reduce antibody binding. To induce antibodies against conserved receptor-binding domain (RBD) regions for protection against SARS-CoV-2 variants of concern and zoonotic sarbecoviruses, we developed mosaic-8b RBD-nanoparticles presenting eight sarbecovirus RBDs arranged randomly on a 60-mer nanoparticle. Mosaic-8b immunizations protected animals from challenges from viruses whose RBDs were matched or mismatched to those on nanoparticles. Here, we describe neutralizing mAbs from mosaic-8b-immunized rabbits, some on par with Pemgarda (the only currently FDA-approved therapeutic mAb). Deep mutational scanning, in vitro selection of spike resistance mutations, and cryo-EM structures of spike-antibody complexes demonstrated targeting of conserved epitopes. Rabbit mAbs included critical D-gene segment features in common with human anti-RBD mAbs, despite rabbit genomes lacking an equivalent human D-gene segment. Thus, mosaic RBD-nanoparticle immunization coupled with multiplexed screening represent an efficient way to generate and select therapeutic pan-sarbecovirus and pan-SARS-2 variant mAbs.

A Nomogram Based on Laboratory Data, Inflammatory Bowel Disease Questionnaire and CT Enterography for Activity Evaluation in Crohn's Disease.

Accurately assessing the activity of Crohn's disease (CD) is crucial for determining prognosis and guiding treatment strategies for CD patients. This study aimed to develop and validate a nomogram for assessing CD activity. The semi-automatic segmentation method and PyRadiomics software were employed to segment and extract radiomics features from the spectral CT enterography images of lesions in 107 CD patients. The radiomic score (rad-score) was calculated using the radiomic signature formula. Multivariate logistic regression analysis identified the independent risk factors of erythrocyte sedimentation rate, fecal calprotectin, and Inflammatory Bowel Disease Questionnaire (IBDQ), and a nomogram was constructed in combination with rad-score. The nomogram underwent evaluation and testing in the training set (n = 84) and validation set (n = 23), respectively. The discrimination performance of the combined (AUC 0.877) was marginally superior to that of IBDQ + clinical (AUC 0.854). However, there was no significant difference in AUC between the two models in the validation set (P = 0.206). IBDQ + clinical outperformed clinical (AUC 0.808), clinical outperformed IBDQ (AUC 0.746), and IBDQ outperformed radiomic signature (AUC 0.688). Significant differences in AUC were observed between the two models (radiomic signature vs clinical, P = 0.026; radiomic signature vs IBDQ + clinical, P = 0.011; radiomic signature vs combined, P = 0.008; in the validation set). The nomogram, combined with laboratory data, IBDQ and rad-score, presents an accurate and reliable method for assessing CD activity. The nomogram enhances the potential for personalized treatment plans and better disease management, making it a valuable tool for clinical practice.

$\ell_1$DecNet+: A New Architecture Framework by $ℓ_1$ Decomposition and Iteration Unfolding for Sparse Feature Segmentation

$\ell_1$DecNet+: A New Architecture Framework by $ℓ_1$ Decomposition and Iteration Unfolding for Sparse Feature Segmentation

FCB-YOLOv8s-Seg: A Malignant Weed Instance Segmentation Model for Targeted Spraying in Soybean Fields

Effective management of malignant weeds is critical to soybean growth. This study focuses on addressing the critical challenges of targeted spraying operations for malignant weeds such as Cirsium setosum, which severely threaten soybean yield in soybean fields. Specifically, this research aims to tackle key issues in plant protection operations, including the precise identification of weeds, the lightweight deployment of segmentation models, real-time requirements for spraying operations, and the generalization ability of models in diverse field environments. To address these challenges, this study proposes an improved weed instance segmentation model based on YOLOv8s-Seg, named FCB-YOLOv8s-Seg, for targeted spraying operations in soybean fields. The FCB-YOLOv8s-Seg model incorporates a lightweight backbone network to accelerate computations and reduce model size, with optimized Squeeze-and-Excitation Networks (SENet) and Bidirectional Feature Pyramid Network (BiFPN) modules integrated into the neck network to enhance weed recognition accuracy. Data collected from real soybean field scenes were used for model training and testing. The results of ablation experiments revealed that the FCB-YOLOv8s-Seg model achieved a mean average precision of 95.18% for bounding box prediction and 96.63% for segmentation, marking an increase of 5.08% and 7.43% over the original YOLOv8s-Seg model. While maintaining a balanced model scale, the object detection and segmentation accuracy of this model surpass other existing classic models such as YOLOv5s-Seg, Mask-RCNN, and YOLACT. The detection results in different scenes show that the FCB-YOLOv8s-Seg model performs well in fine-grained feature segmentation in complex scenes. Compared with several existing classical models, the FCB-YOLOv8s-Seg model demonstrates better performance. Additionally, field tests on plots with varying weed densities and operational speeds indicated an average segmentation rate of 91.30%, which is 6.38% higher than the original model. The proposed algorithm shows higher accuracy and performance in practical field instance segmentation tasks and is expected to provide strong technical support for promoting targeted spray operations.

Semantic segmentation via fusing 2D image and 3D point cloud data with shared multi-layer perceptron

ABSTRACT Three-dimensional (3D) semantic segmentation based on point clouds has become a critical technology in the field of intelligent spatial perception and scene understanding. However, most existing methods cannot fully exploit the spatial feature of 3D point cloud data and the orderliness of two-dimensional (2D) image information, resulting in an inability to effectively improve segmentation accuracy. This study proposes a novel 2D image and 3D point cloud fusion method named shared multi-layer perceptron fusion semantic segmentation (SMFusionSeg) for semantic segmentation. The rich features of 2D image segmentation and spatial information of 3D point clouds are integrated by designing a fusion architecture that facilitates the transfer of 2D semantics to 3D semantics. Specifically, the fusion architecture concatenates the features from both 2D and 3D domains, which are then further integrated through a shared multi-layer perceptron (MLP) and optimized using an attention mechanism to enhance feature effectiveness and distinctiveness. Moreover, a knowledge distillation strategy is introduced to improve the learning capability and segmentation accuracy in parsing complex scenes. Experimental results on the SemanticKITTI dataset demonstrate that the proposed SMFusionSeg achieves a mean Intersection over Union ( mIoU ) of 65.4%, significantly outperforming traditional single-modal methods and many existing segmentation techniques, thereby effectively enhancing the precision and robustness of geographic feature recognition.

An examination of Turkish EFL teacher trainees’ pronunciation knowledge, perception, and production

With the proliferation of research in recent decades, pronunciation has ceased to be the “Cinderella” of language teaching. However, there are still gaps between research findings and classroom implementations (Olson, 2014). To this end, this classroom-based, experimental study explored pronunciation knowledge, perception, and production among first-year Turkish EFL teacher trainees in the context of a sixteen-week-long undergraduate course. Data were collected through three diagnostic tests. In this regard, eighty participants were pretested before and post-tested after the course. The collected data were examined descriptively and inferentially via the IBM SPPS Version 25. Findings indicated dissimilar levels of rise in knowledge (modest), perception (slight), and production (substantial) performances of teacher trainees. Nevertheless, specific gaps in knowledge and problems with segmental and suprasegmental pronunciation features remained. Results were discussed, pedagogical implications were thoroughly assessed, limitations to the study were recognized, and recommendations for future research were made.

Detection of source code vulnerabilities using Nature language processing and deep graph network

The software production sector gains advantages from automated code generating techniques, yet encounters issues related to vulnerabilities in the resulting code. This research presents a hybrid paradigm, termed GBD, for detecting vulnerabilities in software written in C and C++. It integrates Graph Convolution Network (GCN), Bidirectional Encoder Representations from Transformers (BERT), and Dropout. During Phase 2 of the GBD model, the subsequent tasks are executed concurrently: (i) obtaining node and edge features utilizing the GCN graph convolution network; (ii) deriving segment features employing the BERT model; (iii) constructing a source code profile via the Code Property Graph (CPG). Phase 3 of the model implements the Dropout strategy to mitigate overfitting. Phase 4 is the classifier that ascertains the presence of vulnerabilities in the source code. Experimental findings demonstrate the superiority of the proposed model relative to alternative methods, attaining a prediction accuracy of 61.21% for vulnerable code and 88.94% for normal files. Additionally, the classification outcomes demonstrate that with a token length of 512, the GBD model yields the most uniform results across all metrics: Accuracy (86.65%), Precision (38.59%), Recall (66.21%), and F1-score (48.76%). This corresponds with our analysis of the Verum experimental dataset, indicating that over 70% of the source code files have lengths exceeding 256 but less than 512. Furthermore, the GBD model exhibits strong performance across both individual and multiple datasets. For example, in the Verum dataset, the GBD model surpasses five alternative methodologies—REVEAL [1], Russell [2], VulDeePecker [3], SySeVR [4], and Devign [5] - by 4% in Accuracy and between 15% and 57% in Precision, Recall, and F1-score. In comparison to SySeVR [4], the GBD model exceeds it by 3% to 25% across all metrics. In comparison to Devign [5], GBD achieves improvements of 5% to 39% in Precision, Recall, and F1-score. Upon assessment of the FFmpeg+Qume dataset, the GBD model attains an Accuracy improvement ranging from 0.2% to 10% above all other studies. In terms of precision, GBD surpasses alternative methods by 0.3% to 9%. In terms of Recall, GBD is marginally worse than REVEAL by 1.5%, although surpasses all other methodologies by 10% to over 31%. In terms of F1-score, GBD is 0.3% inferior to REVEAL but surpasses other studies by 7% to 30%. The results indicate that the GBD model is effective on both individual and multiple datasets

Classification of diabetic retinopathy grades using CNN feature extraction to segment the lesion

Diabetes's microvascular aftereffect, diabetic retinopathy (DR), is the primary cause of eyesight loss in the globe. In order to prevent vision impairment and to intervene promptly, early detection and precise classification of DR severity are essential. Using standard methods for diagnosing DR requires ophthalmologists to grade cases by hand, a process that can be laborious, subjective, and subject to observer error. In supervised learning task of classification, data instances are classified into predefined classes based on features. The relation between the traits and the classes can be found from the labelled data. After the training is completed, the classes of the unseen data. The frequent reason found for the loss of vision in diabetic retinopathy (DR) is found to be diabetes. Visual damage can be prevented by identifying the degree of DR at right time. For the grading of the DR, deep learning techniques are found to be very effective with maximum possible accuracy. The proposed model is useful in accurately classifying the DR images using the feature extraction with lesion segmentation, by implementing the patterns in the DR images. ReLU activation function is used in the proposed model. CNN feature extraction is used for the important feature extraction by applying the Convolution layers, and edges, textures, and forms are identified. As the model proceeds layer by layer, complicated patterns in the photos can be learned by the model, and can be analysed better. The features of the photos were extracted and found useful in segmentation and classification. ReLU is helpful in improving the convergence and also found useful in learning the patterns. Among the other activation functions, ReLU has higher computational efficiency and therefore is used in the model, which suits well for the DR application. A strong framework is proposed for the classification of the DR grade, for the lesion segmentation and CNN feature extraction. DR categorization using the proposed model is evaluated by data visualization of the important calculated metrics and found to be very effective.

Search anything: segmentation-based similarity search via region prompts

AbstractSearch Anything is presented, a novel approach to perform similarity search in images. In contrast to other approaches to image similarity search, Search Anything enables users to utilize point, box, and text prompts to search for similar regions in a set of images. The region selected by a prompt is automatically segmented, and a binary feature vector is extracted. This feature vector is then used as a query for an image region index, and the images that contain the corresponding regions are returned. Search Anything is trained in a self-supervised manner on mask features extracted by the FastSAM foundation model and semantic features for masked image regions extracted by the CLIP foundation model to learn binary hash code representations for image regions. By coupling these two foundation models, images can be indexed and searched at a more fine-grained level than finding only entire similar images. Experiments on several datasets from different domains in a zero-shot setting demonstrate the benefits of Search Anything as a versatile region-based similarity search approach for images. The efficacy of the approach is further supported by qualitative results. Ablation studies are performed to evaluate how the proposed combination of semantic features and segmentation features together with masking improves the performance of Search Anything over the baseline using CLIP features alone. For large regions, relative improvements of up to 9.87% in mean average precision are achieved. Furthermore, considering context is beneficial for searching small image regions; a context of 3 times an object’s bounding box gives the best results. Finally, we measure computation time and determine storage requirements.

A Method for Detecting Small Target Weld Defects Based on Feature Reorganization Network

Abstract In the process of intelligent weld seam inspection, small weld seams are often prone to false detections or omissions. Existing methods simply concatenate feature segments during the upsampling process without analyzing the importance of each feature segment, leaving room for further improvement in detection accuracy. To address these issues, this study proposes a Feature Reorganization Network (FRNet) for detecting small target weld defects. First, the C2f-Faster-EMA feature extraction module is designed using GSConv convolution, and the LSKNet is introduced to dynamically adjust the receptive field of the backbone in the Backbone section, enhancing the model's ability to extract small target features. Then, a lightweight CARAFE upsampling module is designed in the neck network, which retains more detailed information through feature reorganization and feature expansion, and introduces the parameter-free attention mechanism SimAM to fully capture the contextual information of small targets, thereby enhancing the proposed model's ability to extract small target features. Finally, the GIoU boundary loss function is used to improve the network's bounding box regression performance, achieving intelligent detection of small target weld defects. Experimental results show that the proposed method achieves a mean average precision, parameter count, and computation volume of 85.6%, 2.5M, and 7.0G, respectively, for weld defect detection, outperforming the comparison models and meeting the requirements of practical engineering applications.

Vessel wall MRI evaluation for the safety of endovascular recanalization of non-acute intracranial anterior circulation artery occlusions

BackgroundVessel wall MRI (VWMRI) can reveal the morphological features of intracranial artery occlusion (ICAO). This study aimed to investigate the imaging features of ICAO on VWMRI and explore their correlation...

Semi‐Supervised Medical Image Segmentation Based on Feature Similarity and Multi‐Level Information Fusion Consistency

ABSTRACTSemantic segmentation is a key task in computer vision, with medical image segmentation as a prominent downstream application that has seen significant advancements in recent years. However, the challenge of requiring extensive annotations in medical image segmentation remains exceedingly difficult. In addressing this issue, semi‐supervised semantic segmentation has emerged as a new approach to mitigate annotation burdens. Nonetheless, existing methods in semi‐supervised medical image segmentation still face challenges in fully exploiting unlabeled data and efficiently integrating labeled and unlabeled data. Therefore, this paper proposes a novel network model—feature similarity multilevel information fusion network (FSMIFNet). First, the feature similarity module is introduced to harness deep feature similarity among unlabeled images, predicting true label constraints and guiding segmentation features with deep feature relationships. This approach fully exploits deep feature information from unlabeled data. Second, the multilevel information fusion framework integrates labeled and unlabeled data to enhance segmentation quality in unlabeled images, ensuring consistency between original and feature maps for comprehensive optimization of detail and global information. In the ACDC dataset, our method achieves an mDice of 0.684 with 5% labeled data, 0.873 with 10%, 0.884 with 20%, and 0.897 with 50%. Experimental results demonstrate the effectiveness of FSMIFNet in semi‐supervised semantic segmentation of medical images, outperforming existing methods on public benchmark datasets. The code and models are available at https://github.com/liujiayin12/FSMIFNet.git.

A Convolutional Graph Neural Network Model for Water Distribution Network Leakage Detection Based on Segment Feature Fusion Strategy

In this study, an innovative leak detection model based on Convolutional Graph Neural Networks (CGNNs) is proposed to enhance response speed during pipeline bursts and to improve detection accuracy. By integrating node features into pipe segment features, the model effectively combines CGNN with water distribution networks, achieving leak detection at the pipe segment level. Optimizing the receptive field and convolutional layers ensures high detection performance even with sparse monitoring device density. Applied to two representative water distribution networks in City H, China, the model was trained on synthetic leak data generated by EPANET simulations and validated using real-world leak events. The experimental results show that the model achieves 90.28% accuracy in high-density monitoring areas, and over 85% accuracy within three pipe segments of actual leaks in low-density areas (10%-20%). The impact of feature engineering on model performance is also analyzed and strategies are suggested for optimizing monitoring point placement, further improving detection efficiency. This research provides valuable technical support for the intelligent management of water distribution networks under resource-limited conditions.

UnifiedCut: A Simple and Efficient Neural Model for Thai, Burmese and Khmer Word Segmentation

Word segmentation is a critical task in natural language processing for southeast Asian Abugida languages, including Thai, Burmese, and Khmer. Existing approaches demonstrate that models using fixed-length windowed context inputs can achieve high segmentation accuracy; however, they often rely on low-level character features or language-specific preprocessing. Character-based methods can limit feature learning, while language-specific features add complexity due to specialized preprocessing requirements. This paper introduces UnifiedCut, which is a neural model that leverages multiple n-grams within a windowed multi-head attention mechanism. This design captures segmentation features from local contexts and multi-perspective n-gram inputs, enhancing generalization and recall, particularly for out-of-vocabulary words. Compared to CNN- and RNN-based approaches, UnifiedCut’s multi-head attention enables finer-grained feature extraction and greater parallelism, resulting in a faster, more scalable solution. Comprehensive experiments on public datasets for Thai, Burmese, and Khmer show that UnifiedCutachieves state-of-the-art performance in word segmentation.

Identifying vehicle types from trajectory data based on spatial-semantic information

ABSTRACT Obtaining information about various vehicle types traveling on road networks is crucial for estimating traffic loads on roads, evaluating the adequacy of road design standards, and providing personalized navigation guidance. Traditional intrusive and non-intrusive methods for vehicle-type identification often encounter challenges such as high maintenance costs, incomplete coverage of all roads, and technical limitations in adverse weather conditions. In recent years, vehicle GNSS trajectory data accumulation has provided a continuous, dynamic, wide-coverage, and cost-effective data resource for identifying vehicle types. However, existing trajectory-based methods still have some drawbacks in micro-trip segmentation and limitation of movement features. Hence, this paper proposes a novel approach for identifying vehicle types by leveraging spatial-semantic information from trajectory data. This proposed method initially detects staying points from trajectory data, then utilizes DBSCAN clustering on these detected staying points to adaptively segment original vehicle trajectories into various micro-trips in a steady-moving pattern. Subsequently, several statistical indicators related to velocity and acceleration are calculated as movement features for each micro-trip. Additionally, each vehicle trajectory’s driving-road hierarchy and staying-place information are quantified as the geo-semantic features. Finally, the calculated movement and geo-semantic features are utilized for two classification tasks using three typical classification models. Experimental results demonstrate that the proposed method achieves reliable performance in identifying various vehicle types by incorporating adaptive micro-trip segmentation and multiple spatial-semantic features, particularly outperforming fixed-size trip segmentation and using only movement features. Furthermore, it is observed that the classification accuracy of coaches, trucks, and semis is consistently higher than that for large, medium, and small passenger cars, indicating that the vehicle purposes may be more distinguishable than vehicle loads.

TranSenseFusers: A temporal CNN-Transformer neural network family for explainable PPG-based stress detection

Stress is a common everyday emotional state in modern society contributing to both physical and mental illnesses. Thus, detecting and managing the degree of stress is crucial to improve well-being. Wearable devices equipped with biosensors, such as PhotoPlethysmoGraphy (PPG), can measure reliably a person’s affective state. However, PPG-based approaches suffer from the presence of Motion Artifacts (MA) affecting their overall performance. Classical machine learning and deep learning approaches have been proposed over the years for PPG-based stress detection, exploiting signal processing techniques to remove the recorded noise, but lack explainability or their performance fails to generalize across subjects. In the current work, we present a novel architecture, TranSenseFuser comprised of temporal convolutions followed by feature-level or sequence-level multi-head attention to improve sensor fusion’s effectiveness and exploit the provided attention maps as a form of explainability. The developed models are evaluated on highly benchmarked public dataset, namely WESAD, achieving state-of-the-art results (up to 98.46% accuracy and 97.03% F1-score) using different window sizes and cross-validation set-ups. Moreover, we demonstrate the explainability of the model towards filtering out the motion artifacts by visualizing the obtained attention maps and quantify the performance of this artifact segmentation feature in a zeros-shot manner.

Workflow automation of SEM acquisitions and feature tracking.

Acquiring multiple high magnification, high resolution images with scanning electron microscopes (SEMs) for quantitative analysis is a time consuming and repetitive task for microscopists. We propose a workflow to automate SEM image acquisition and demonstrate its use in the context of nanoparticle (NP) analysis. Acquiring multiple images of this type of specimen is necessary to obtain a complete and proper characterization of the NP population and obtain statistically representative results. Indeed, a single high magnification image only scans a small area of sample, containing only few NPs. The proposed workflow is successfully applied to obtain size distributions from image montages at three different magnifications (20,000x, 60,000x and 200,000x) on the same area of the sample using a Python based script. The automated workflow consists of sequential repositioning of the electron beam, stitching of adjacent images, feature segmentation, and NP size computation. Results show that NPs are best characterized at higher magnifications, since lower magnifications are limited by their pixel size. Increased accuracy of feature characterization at high magnification highlights the importance of automation: many high-magnification acquisitions are required to cover a similar area of the sample at low magnification. Therefore, we also present feature tracking with smart beam positioning as an alternative to blind acquisition of very large image arrays. Feature tracking is achieved by integrating microscope tasks with image processing tasks, and only areas of interest will be imaged at high resolution, reducing total acquisition duration.

Machine Learning for Automated Oil Palm Fruit Grading: The Role of Fuzzy C-Means Segmentation and Textural Features

Oil palm fruit, a high-demand and economically valuable crop, faces grading challenges in Malaysia due to labour-intensive methods, resulting in unharvested ripe fruits and yield losses. Extensive research has been conducted on outdoor palm fruit classifiers, but most high-accuracy research in this field relies heavily on large image datasets. This creates a trade-off between employing more sophisticated deep-learning approaches that require extensive data and utilising traditional techniques that have lower data requirements. Therefore, the aim is to propose an outdoor image-based fresh fruit bunch (FFB) classification system, emphasising pre-processing technique and feature extraction techniques suitable for limited datasets. To achieve this, the FFB was localised within the original images using the YOLOv4-Tiny algorithm. Subsequently, a salient segmentation method utilising superpixel-based Fast Fuzzy C-means (FFCM) clustering is employed to remove the background from the images. Next, colour moment and opposite colour local binary pattern (OCLBP) features are extracted from the segmented images to capture important information for classification. Finally, the extracted features are fed into a Multilayer Perceptron (MLP) classifier, which enables the system to predict five classes: damaged, empty, unripe, ripe, and overripe. The developed system demonstrated a commendable performance in accurately classifying fruit bunches, achieving an accuracy of 93.68%. In conclusion, the proposed system effectively addresses the issue of unripe fruit harvesting and contributes to the advancement of state-of-the-art methods in classifying outdoor FFB images.