Accurate Classification Results Research Articles

Efficient UAV-Based Automatic Classification of Cassava Fields Using K-Means and Spectral Trend Analysis

High-resolution images captured by Unmanned Aerial Vehicles (UAVs) play a vital role in precision agriculture, particularly in evaluating crop health and detecting weeds. However, the detailed pixel information in these images makes classification a time-consuming and resource-intensive process. Despite these challenges, UAV imagery is increasingly utilized for various agricultural classification tasks. This study introduces an automatic classification method designed to streamline the process, specifically targeting cassava plants, weeds, and soil classification. The approach combines K-means unsupervised classification with spectral trend-based labeling, significantly reducing the need for manual intervention. The method ensures reliable and accurate classification results by leveraging color indices derived from RGB data and applying mean-shift filtering parameters. Key findings reveal that the combination of the blue (B) channel, Visible Atmospherically Resistant Index (VARI), and color index (CI) with filtering parameters, including a spatial radius (sp) = 5 and a color radius (sr) = 10, effectively differentiates soil from vegetation. Notably, using the green (G) channel, excess red (ExR), and excess green (ExG) with filtering parameters (sp = 10, sr = 20) successfully distinguishes cassava from weeds. The classification maps generated by this method achieved high kappa coefficients of 0.96, with accuracy levels comparable to supervised methods like Random Forest classification. This technique offers significant reductions in processing time compared to traditional methods and does not require training data, making it adaptable to different cassava fields captured by various UAV-mounted optical sensors. Ultimately, the proposed classification process minimizes manual intervention by incorporating efficient pre-processing steps into the classification workflow, making it a valuable tool for precision agriculture.

Crop classification in the middle reaches of the Hei River based on model transfer

Crop classification using remote sensing technology is highly important for monitoring agricultural resources and managing water usage, especially in water-scarce regions like the Hei River. Crop classification requires a substantial number of labeled samples, but the collection of labeled samples demands significant resources and sample data may not be available for some years. To classify crops in sample-free years in the middle reaches of the Hei River, we generated multisource spectral data (MSSD) based on a spectral library and sample data. We pre-trained a model using labeled samples, followed by fine-tuning the model with MSSD to complete the crop classification for the years without samples. We conduct experiments using three CNN-based deep learning models and a machine learning model (RF). The experimental results indicate that in the model transfer experiments, using a fine-tuned model yields accurate classification results, with overall accuracy exceeding 90%. When the amount of labeled sample data is limited, fine-tuning the model based on MSSD can enhance the accuracy of crop classification. Overall, fine-tuning models based on MSSD can significantly enhance the accuracy of model transfer and reduce the reliance of deep learning models on large-scale sample datasets. The method to classify crops in the middle reaches of the Hei River can provide data support for local resource utilization and policy formulation.

Detection of Diffusion Interlayers in Dissimilar Welded Joints in Processing Pipelines by Acoustic Emission Method

The paper considers the neural network application to detect microstructure defects in dissimilar welded joints using the acoustic emission (AE) method. The peculiarity of the proposed approach is that defect detection is carried out taking into account a priori information about the properties of the AE source and the acoustic waveguide parameters of the testing structure. Industrial process pipelines with dissimilar welded joints were studied as the testing object, and diffusion interlayers formed in fusion zones of welded joints were considered microstructure defects. The simulation of AE signals was carried out using a hybrid method: the signal waveform was determined based on a finite element model, while the amplitudes of AE hits were determined based on a physical experiment on mechanical testing of dissimilar welded joints. Measurement data from industrial process pipelines were used as noise realizations. As a result, a data sample was formed that considered the parameters of the AE source and the parameters of the acoustic waveguide with realistic noise parameters and a signal-to-noise ratio. The proposed method allows for a more accurate determination of the waveform, spectrum, and amplitude parameters of the AE signal. Greater certainty in the useful signal parameters allows for achieving a more accurate and reliable classification result. When using a backpropagation neural network, a percentage of correct classification of more than 90% was obtained for a data set in which the signal-to-noise ratio was less than (−5 dB) in 90% of cases.

Comparative analysis of X-ray image classification of pneumonia based on deep learning algorithm

Abstract. In this paper, we compare and analyse the effectiveness of deep learning models such as AlexNet, Vgg, GoogleNet and MobileNet in the task of pneumonia image classification. During the training process, GoogleNet showed the fastest convergence speed, reaching convergence at the 3rd epoch; subsequently, the other three models gradually stabilised. In the end, the loss of AlexNet is 0.426, the loss of Vgg is 0.566, the loss of GoogleNet is 0.936, and the loss of MobileNet is 0.626. By selecting the weights of the round with the best training effect of each model, the results of classification accuracy are obtained: 88.9% for AlexNet, 92.6% for Vgg, and 92.6% for GoogleNet. 92.6%, GoogleNet is 85.2%, and MobileNet is 96.3%. MobileNet demonstrated the best prediction performance on the test set. These results provide a useful reference for the application of deep learning models in medical imaging..

Numerical parametric study for ballast assessment using SVM applied to GPR data

Abstract This study explores novel radar-based methodologies for ballast fouling evaluation and structural anomalies in railway structures. Ground Penetrating Radar (GPR), coupled with numerical modeling, is utilized to provide insights into ballast condition and structure. Various data processing methods, including the Matrix Pencil Method (MPM) and Full Waveform Inversion (FWI), are investigated for their effectiveness in detecting fouling. Support Vector Machine (SVM)-based machine learning approaches are employed to enhance classification accuracy. The results demonstrate that integrating raw GPR signals with MPM yields the most accurate classification results, facilitating efficient assessment of ballast condition and contributing to improved railway maintenance strategies.

Detection of clavicle fracture after martial arts training based on 3D segmentation and multi-perspective fusion

To improve the diagnosis efficiency of clavicle fractures, a Diagnosis Model of Clavicle Fracture Based on 3D Segmentation and Multi Perspective Fusion (3Ds MPF) is proposed in the experiment. By performing 3D segmentation on clavicle images, key layer images are extracted, and multi-perspective image data are fused and output to achieve accurate classification and diagnosis results. The results show that on the relevant dataset, the loss function value of the constructed model is less than 0.01, and the calculation accuracy is as high as 0.999. The average classification accuracy in actual use exceeds 90%, and it can capture the majority of fracture features in clavicle computed tomography images. The above results indicate that the model constructed by the research institute can significantly improve the judgment efficiency of computed tomography images, which helps to enhance the detection and diagnosis of clinical fracture conditions. The experimental results have helped improve the clinical diagnostic efficiency of fracture detection work.

Enhanced QSVM with elitist non‐dominated sorting genetic optimisation algorithm for breast cancer diagnosis

AbstractMachine learning has emerged as a promising method for predicting breast cancer using quantum computation techniques. Quantum machine learning algorithms, such as quantum support vector machines (QSVMs), are demonstrating superior efficiency and economy in tackling complex problems compared to traditional machine learning methods. When compared with classical support vector machine, the quantum machine produces remarkably accurate results. The suggested quantum SVM model in this study effectively resolved the binary classification problem for diagnosing malignant breast cancer. This work introduces an enhanced approach to breast cancer diagnosis by integrating QSVM with elitist non‐dominated sorting genetic optimization (ENSGA), leveraging the strengths of both techniques to achieve more accurate and efficient classification results. ENSGA plays a crucial role in optimising QSVM parameters, ensuring that the model attains the best possible classification accuracy while considering multiple objectives simultaneously. Moreover, the quantum kernel estimation method demonstrated exceptional performance by achieving high accuracy within an impressive execution time of 0.14 in the IBM QSVM simulator. The seamless integration of quantum computation techniques with optimisation strategies such as ENSGA highlights the potential of quantum machine learning in revolutionising the field of healthcare, particularly in the domain of breast cancer diagnosis.

Lightweight skin cancer detection IP hardware implementation using cycle expansion and optimal computation arrays methods

Skin cancer is recognized as one of the most perilous diseases globally. In the field of medical image classification, precise identification of early-stage skin lesions is imperative for accurate diagnosis. However, deploying these algorithms on low-cost devices and attaining high-efficiency operation with minimal energy consumption poses a formidable challenge due to their intricate computational demands. This study proposes a lightweight hardware design based on a convolutional neural network (CNN) for real-time processing of skin disease classifiers on portable devices. Our skin cancer recognition processor utilizes an optimally parallel designed processing engine (PE) for global computation, which greatly reduces hardware resource utilization by multiplexing of computational unit circuits. In addition, a design approach that provides loop unrolling effectively reduces the number of data accesses, thereby reducing computational complexity and logic resource requirements. The hardware circuits in this design perform data inference in convolutional, pooling, and fully connected layers based on 16-bit floating-point numbers. Evaluation of the HAM10000 database dataset shows that the architecture achieves an average classification accuracy of 97.8 %. We are the first to implement an all-hardware FPGA-based skin cancer detection platform that offers a 3.5x speedup in recognition compared to existing skin cancer accelerators at 50 MHz while consuming only 0.48 W of power. The implementation of this hardware architecture meets the major constraints of portable devices, featuring low resource utilization, low power consumption, and cost-effectiveness, while still providing efficient classification and high accuracy results.

An evaluation of single and multi-date Landsat image classifications using random forest algorithm in a semi-arid savanna of Ghana, West Africa

Accurate detection and quantification of land use and land cover (LULC) change is critical for understanding landscape patterns in heterogeneous semi-arid environments. This study investigates the performance of single-date and multi-date Landsat images as well as the relationship between different LULC schemes (simple [2 and 4 classes] and complex [6 and 9 classes]) and the resulting classification accuracy. Specifically, the random forest algorithm was applied to Landsat data comprised of different combinations of image dates (single-date and multi-date) captured in June, October, and December for multiple levels of LULC (scheme) mapping and accuracy evaluations due to its high performance when dealing with large data and heterogeneous landscapes. Results indicated that multi-date images consistently produced higher classification accuracies than single-date images. Significant negative correlations observed between the number of classes in LULC schemes and overall accuracy and kappa coefficient indicate that the more complex the LULC scheme, the lower the accuracy produced. Nevertheless, improvement in overall accuracy was negligible for simple schemes (e.g., ∼1 % for two LULC classes), while it was moderate for complex schemes (∼5 %) when using the best-performing images for multi-date (June-October-December) compared to single-date (October) classifications: however, the improvement was considerable when compared to the least performing single-date image (June, 8–15 %). These varying classification accuracies were due to differences or similarities in spectral responses of target classes in the various LULC schemes applied to the investigated images. Consequently, the resulting differences in the spatial distribution and quantification of LULC classes produced by the different approaches can affect policy and land management decisions, especially if inappropriate image dates are used for LULC mapping. Overall, the findings highlight the reliability of appropriate single-date and multi-date images for mapping LULC change using simple and complex schemes in heterogeneous semi-arid savanna landscapes.

A fault diagnosis method for bearings and gears in rotating machinery based on data fusion and transfer learning

Abstract Rotating machinery is a crucial component of industrial equipment, and the fault diagnosis of bearings and gears, as vital elements of rotating machinery, is essential since they often fail under harsh working conditions, leading to significant property losses and serious personal safety problems. However, fault data for gears and bearings are often sparse in actual condition, and it is a challenge to ensure the reliability and stability of fault diagnosis results by extracting the features of a single data. To solve the above problems, this paper proposes a fault diagnosis method that combines Transfer Learning and data fusion techniques. Firstly, in this method, two kinds of fault signals are transformed into Gramian Angular Difference Fields and Recurrence Plot. Next, a U-shaped feature fusion dual discriminator generative adversarial network is used to fuse two-dimensional images from multiple sensor data. Its feature fusion module deeply integrates the features of the two images, thereby solving the impact of single data on the reliability and stability of fault diagnosis. Moreover, open-source datasets are used for Transfer Learning training to tackle the small sample problem. Finally, a decision-level information fusion classifier, the Dual-Branch Dempster-Shafer Classifier (DB-DSC), classifies the fused images. This classifier incorporates an improved soft threshold function and D-S evidence theory to achieve adaptive gradient changes and improve the robustness and accuracy of classification results. The experimental results show the effectiveness and stability of the proposed method, and the generated images get high score in several metrics. The average classification accuracy of the classification network reaches 93% and 92.5% on the two datasets, Therefore, the proposed method exhibits strong fault diagnosis capabilities under the small sample conditions of bearings and gears.

A multi-slice attention fusion and multi-view personalized fusion lightweight network for Alzheimer’s disease diagnosis

ObjectiveAlzheimer’s disease (AD) is a type of neurological illness that significantly impacts individuals’ daily lives. In the intelligent diagnosis of AD, 3D networks require larger computational resources and storage space for training the models, leading to increased model complexity and training time. On the other hand, 2D slices analysis may overlook the 3D structural information of MRI and can result in information loss.ApproachWe propose a multi-slice attention fusion and multi-view personalized fusion lightweight network for automated AD diagnosis. It incorporates a multi-branch lightweight backbone to extract features from sagittal, axial, and coronal view of MRI, respectively. In addition, we introduce a novel multi-slice attention fusion module, which utilizes a combination of global and local channel attention mechanism to ensure consistent classification across multiple slices. Additionally, a multi-view personalized fusion module is tailored to assign appropriate weights to the three views, taking into account the varying significance of each view in achieving accurate classification results. To enhance the performance of the multi-view personalized fusion module, we utilize a label consistency loss to guide the model’s learning process. This encourages the acquisition of more consistent and stable representations across all three views.Main resultsThe suggested strategy is efficient in lowering the number of parameters and FLOPs, with only 3.75M and 4.45G respectively, and accuracy improved by 10.5% to 14% in three tasks. Moreover, in the classification tasks of AD vs. CN, AD vs. MCI and MCI vs. CN, the accuracy of the proposed method is 95.63%, 86.88% and 85.00%, respectively, which is superior to the existing methods.ConclusionsThe results show that the proposed approach not only excels in resource utilization, but also significantly outperforms the four comparison methods in terms of accuracy and sensitivity, particularly in detecting early-stage AD lesions. It can precisely capture and accurately identify subtle brain lesions, providing crucial technical support for early intervention and treatment.

A multi-lingual sign language recognition system using machine learning

AbstractRecently, automatic sign language recognition field gets a great attention. Machine learning algorithms are mainly used to recognize hand gestures. Most of recent studies train their machine learning model using a specific sign language of a specific country such as the American Sign Language. In this paper, we propose a multi-lingual sign language system based machine learning that is called Multi-lingual Sign Languages Interpreter (MSLI) system. MSLI trains a machine learning model based on hand signs of multiple languages. It can detect the language of the input signs and their labels. In a case of input testing signs with the same language, the proposed system can provide two-steps recognition, where it only detects the language of the first sign, and then the rest signs are tested according to the recognized language. Also, MSLI can provide separate classification of signs per each language. Experiments were performed using 11 datasets with different languages. Separate and combined classification was performed on the input data. Experimental results show the accuracy of the proposed system. Training accuracy of the proposed system over most of the used separate different sign language datasets is approximately ranged from 90 to 100%. Also, most classification accuracy results of the test data of the separate datasets exceeded 90%. The combined classification of proposed MSLI archived training accuracy of 95.87% and testing accuracy of 92.33%.

Using deep learning and pretreatment EEG to predict response to sertraline, bupropion, and placebo

ObjectivePredicting an individual’s response to antidepressant medication remains one of the most challenging tasks in the treatment of major depressive disorder (MDD). Our objective was to use the large EMBARC study database to develop an electroencephalography (EEG)-based method to predict response to antidepressant treatment. MethodsPre-treatment EEG data were collected from study participants treated with either sertraline (N = 105), placebo (N = 119), or bupropion (N = 35). After preprocessing, the robust exact low-resolution electromagnetic tomography (ReLORETA) brain source localization method was used to reconstruct the source signals in 54 brain regions. Connectivity between regions was determined using symbolic transfer entropy (STE). A convolutional neural network (CNN) classified participants as responders or non-responders to each treatment. ResultsClassification accuracy was 91.0%, 95.4%, and 86.8% for sertraline, placebo, and bupropion, respectively. The most highly predictive features were connectivity between i) the anterior cingulate cortex and superior parietal lobule (alpha frequency), ii) the anterior cingulate cortex and orbitofrontal area (beta frequency), and iii) the orbitofrontal area and anterior cingulate cortex (gamma frequency). ConclusionCNN analysis of EEG connectivity may accurately predict response to sertraline, bupropion, and placebo. SignificanceThe suggested method may offer clinicians an accessible and cost-effective tool for speedy treatment and helps pharmaceutical firms to test new antidepressants efficiently.

Classification method of seabed sonar image substrate based on ELM-AdaBoost

The results of sediment classification are affected by the measurement environment. The water depth has a significant impact on the sound propagation, and the acoustic characteristics will be different under different water depths. In addition, seabed topography, sediment types and their distribution, acoustic frequency, equipment types and settings, and suspended substances in water will all affect the original sonar data. In order to obtain accurate classification results, a seabed sonar image classification method based on Elm AdaBoost is proposed. The original sound intensity data is compensated for gain and the anomaly is eliminated. After preprocessing the multi beam sonar data, the image texture feature is extracted from the gray level co-occurrence matrix as the feature vector for classification, and the elm classifier is constructed. The elm is enhanced by AdaBoost. The calculated feature is input into the elm AdaBoost network for sediment classification, and the result is output. Simulation results show that the proposed method effectively improves the accuracy of sediment classification, and verifies the feasibility of this method.

Multi-level interactive fusion network based on adversarial learning for fusion classification of hyperspectral and LiDAR data

With the growing awareness of the limitations of unimodal data, research on joint classification using multimodal remote sensing data has garnered increasing attention. However, existing methods still have certain deficiencies in dealing with feature extraction and feature fusion of multimodal remote sensing data. In order to obtain more accurate classification results, it is crucial to fully exploit and utilize the complementary information in multimodal remote sensing data. In this paper, a multi-level interactive fusion network based on adversarial learning (MLIF-AL) is proposed for hyperspectral image (HSI) and light detection and ranging (LiDAR) data fusion classification. First, the feature extraction part based on generative adversarial network (GAN) utilizes adversarial training between the generator and the discriminator to guide the network to learn more discriminative and distinguished feature representations. Meanwhile, a cross-modal interactive information extraction (CMIIE) module is also designed for the generator encoding part to promote the interaction between multimodal data from a global perspective and realize the full utilization of multimodal complementary information. In addition, in the multi-level feature fusion classification (MLFFC) part, the complementarities between the features learned by the neural networks at each layer are comprehensively utilized to obtain higher-level multimodal fusion semantic information. Finally, the adversarial loss and classification loss are combined for network training. Extensive experiments on three commonly used HSI and LiDAR datasets and comparisons with state-of-the-art methods demonstrate the effectiveness and superiority of the model.

Comparison of Multiple Machine Learning Models for the Classification of Cell States Based on Impedance Features

Abstract Microfluidic impedance flow cytometry (IFC) enables high-throughput single-cell analysis in label-free manner. Tens of thousands of cells can be measured in several minutes under multiple frequencies, which give rise to impedance features with rich information ideal for machine learning (ML)-based cell classification. Conventional data processing approach for IFC typically exploits the scattered distribution of the measured cells which correlates the impedance features (e.g., the impedance amplitude and phase at different frequencies, the amplitude ratio between high to low frequencies) and exhibits resolved cell clusters in scatter plot. By manually gating on the distributed dots plot, the cell subgroups get mapped to different cell type or cellular states. ML-based data processing for IFC not only reduces the human workload, and more importantly, it also eliminates the human interference to manual gating strategy, and thus potentially leading to more concise and accurate cell classification results. Here, we demonstrate the ML-based classification of different cell states for tumor cells subject to anticancer drug treatment. IFC-measured impedance data of H1650 cells and Hela cells under drug-induced mitosis block state and apoptosis state have been applied for ML-based cell state identification. Three machine learning models, including the random forest (RF), support vector machine (SVM) and K-nearest neighbours (KNN) have been trained for impedance features extracted from cell signals at both 500 kHz and 10 MHz. In comparison, the RF model give rise to the highest classification accuracies among all trained models here. For H1650 cells, 84.01% and 85.96% accuracies have been respectively achieved for G1/S state vs. apoptosis and G2/M vs. apoptosis. For the classification between G2/M vs. apoptosis for the paclitaxel-treated Hela cells, the RF model produces high accuracy of 98.70%.

Machine learning approach to burned area mapping for Southern California

ABSTRACT Accurate representation of the location and amount of burned areas is vital to the understanding of patterns and impacts of fires. Some extant burned area maps appear to have high commission errors, which lead to an overrepresentation of burned area. The primary research objective of this study was to assess whether region-specific training data used for machine learning routines improve accuracy of burned area products for western San Diego County. We used training data derived from fine-scale aerial orthoimagery to create and compare three training sets, each with a different Landsat scale sub-pixel burn threshold: 20%, 50%, or 80%. Meaning either 20%, 50%, or 80% of a 30 m × 30 m pixel had to burn for the entire pixel to be classified as burned. High-resolution orthoimagery was used for the creation of the training/testing data as well as determining which sub-pixel threshold leads to more accurate burned area representation. These training data were input into a gradient-boosted regression model. We compared the burned area product from the region-specific gradient boosted model (L-GBRM) to the three products: Monitoring Trends in Burn Severity, Fire and Resource Assessment Program, and the Landsat Burned Area product. We found >20% sub-pixel burn threshold of a Landsat pixel yielded the most accurate classification results. We used a 50% sub-pixel burn threshold for the reference data to compare the results to since the burned area associated with it is closely aligned with the high-resolution orthoimagery determined burn area. The L-GBRM was the most accurate product while also mapping the smallest area burned, suggesting that the extant products have relatively high commission errors. Using region-specific training data achieved a higher accuracy than nationwide training data. Looking at sub-pixel burn thresholds for creating a burned area map could prove to make a more accurate map in terms of area burned represented.

Data homogeneity impact in tree species classification based on Sentinel-2 multitemporal data case study in central Sweden

ABSTRACT Spatial information on forest composition is invaluable for achieving scientific, ecological, and management objectives and for monitoring multiple changes in forest ecosystems. The increased flow of optical satellite data provides new opportunities to improve tree species mapping. However, the accuracy of such maps is affected by training data, and in particular on the homogeneity of individual classes. Thus, we evaluated the effect of data homogeneity in tree species classification. We performed tree species classification by considering different ways to partition data into tree species classes. The class sets considered were (i) only mixed coniferous and mixed deciduous forest classes, (ii) single-species classes, (iii) single-species, mixed coniferous and mixed deciduous classes, and (iv) single-species, mixed coniferous and mixed deciduous classes and a true mixed class. Using data from the Swedish National Forest Inventory, we varied the threshold that defined dominating species. Tree species were classified for a study area in central Sweden using Sentinel-2 data and two classification approaches: Bayesian inference and random forest (RF). Images were selected by class separability and the most informative images based on variable selection with RF. The most informative images tended to be selected by both methods. However, in forests with tree species of similar spectral behaviour, image selection on the basis of class separability was found to be more reliable. More accurate classification results were achieved as the number of classes decreased and the threshold of plot purity increased. The Bayesian classification approach of only mixed coniferous and mixed deciduous classes gave the highest OA, always greater than 90%. When discriminating between pure plots of Birch (Betula spp.), Spruce (Picea abies), Scots pine (Pinus sylvestris) and Lodgepole pine (Pinus contorta), the best OA values were 84% for Bayesian and 80% for RF. In more complicated scenarios, RF resulted in higher overall accuracies (OA).

Deep and hand-crafted features based on Weierstrass elliptic function for MRI brain tumor classification

Abstract Advances in medical imaging and artificial intelligence have led to improvements in diagnosis and non-invasive patient examination accuracy. The use of the fundamental method for Magnetic resonance imaging (MRI) brain scans as a screening tool has increased in recent years. Numerous studies have proposed a variety of feature extraction methods to classify the abnormal growths in MRI scans. Recently, the MRI texture analysis and the use of deep features have resulted in remarkable performance improvements in the classification and diagnosis of challenging pathologies, like brain tumors. This study proposes employing a handcrafted model based on Weierstrass elliptic function (WEF) and deep feature based on DenseNet-201 to classify brain tumors in MRI images. By calculating the energy of each individual pixel, the Weierstrass coefficients of the WEF are used to capture high frequency image details of the brain image. The WEF mode works to extract the nonlinear patterns in MRI images based on the probability of each pixel. While the dense connectivity of DenseNet-201’s architecture allows to learn features at multiple scales and abstraction levels. These features are passed to support vector machines classifier, which classifies the brain tumor. The results of classification accuracy achieved is 98.55% for combined features of WEF with trained DenseNet-201. Findings on the brain tumor segmentation dataset indicated that the proposed method performed better than alternative techniques for classifying brain tumors.

A stereoscopic video computer vision system for weed discrimination in rice field under both natural and controlled light conditions by machine learning

A site-specific weed detection and classification system was implemented with a stereoscopic video camera to reduce the adverse effects of chemical herbicides in rice field. A computer vision and meta-heuristic hybrid NN-ICA classifier were used to accurately discriminate between two weed varieties and rice plants, under either natural light (NLC) or controlled light conditions (CLC). Preprocessing, segmentation, and matching procedures were performed on images coming from either right or left camera channels. Most discriminant features were selected from average, either arithmetic or geometric, images using a NN-PSO algorithm. Accuracy classification results with the stereo computer vision system under NLC were 85.71 % for the arithmetic mean (AM) and 85.63 % for the geometric mean (GM), test set. At the same time, accuracy classification results of the computer vision system under CLC reached 96.95 % for the AM case and 94.74 % for the GM case, being consistently higher than those under NLC.