True Positive Detection Rate Research Articles

RGB Contrast 영상에서의 Local Binary Pattern Variance를 이용한 연기검출 방법

Smoke detection plays an important role for the early detection of fire. In this paper, we suggest a newly developed method that generated LBPV(Local Binary Pattern Variance)s as special feature vectors from RGB contrast images can be applied to detect smoke using SVM(Support Vector Machine). The proposed method rearranges mean value of the block from each R, G, B channel and its intensity of the mean value. Additionally, it generates RGB contrast image which indicates each RGB channel’s contrast via smoke’s achromatic color. Uniform LBPV, Rotation-Invariance LBPV, Rotation-Invariance Uniform LBPV are applied to RGB Contrast images so that it could generate feature vector from the form of LBP. It helps to distinguish between smoke and non smoke area through SVM. Experimental results show that true positive detection rate is similar but false positive detection rate has been improved, although the proposed method reduced numbers of feature vector in half comparing with the existing method with LBP and LBPV.

Stochastic model: reCAPTCHA controller based co-variance matrix analysis on frequency distribution using trust evaluation and re-eval by Aumann agreement theorem against DDoS attack in MANET

The DDoS attack is more vulnerable attack in the self structured and infra structure less MANET environment. The bot net based flooding attack is very cheap and easy to implement. Many detection and prevention mechanisms has been proposed to avoid DDoS attack. Here we have proposed the stochastic model of covariance matrix analysis in which the entropy difference has been computed by the frequency distribution of rule metrics against the threshold entropy periodically which in turn difference is highest can be controlled by reCAPTCHA controller to defend against DDoS attack. The prevention mechanism by trust evaluation using clustering methodology used to identify the attacker and reduce the cost of detection mechanism. The Aumann agreement theorem based trust re-evaluation is also proposed in order to reduce the false positive and negative probabilities, such that the accuracy of the system is enhanced. Finally the nodes identified as attacker by their trust values will be evicted and rekeying has been done. The experimental analysis in NS2 with CAIDA dataset shows true positive rate and detection rate is high in the proposed mechanism.

Next-generation sequencing and its initial applications for molecular diagnosis of systemic auto-inflammatory diseases

ObjectivesSystemic auto-inflammatory disorders (SAIDs) are a heterogeneous group of monogenic diseases sharing a primary dysfunction of the innate immune system. More than 50% of patients with SAID does not show...

Improving Reliability of Monitoring Background EEG Dynamics in Asphyxiated Infants.

The goal of this study is to develop an automated algorithm to quantify background electroencephalography (EEG) dynamics in term neonates with hypoxic ischemic encephalopathy. The recorded EEG signal is adaptively segmented and the segments with low amplitudes are detected. Next, depending on the spatial distribution of the low-amplitude segments, the first part of the algorithm detects (dynamic) interburst intervals (dIBIs) and performs well on the relatively artifact-free EEG periods and well-defined burst-suppression EEG periods. However, on testing the algorithm on EEG recordings of more than 48h per neonate, a significant number of misclassified and dubious detections were encountered. Therefore, as the next step, we applied machine learning classifiers to differentiate between definite dIBI detections and misclassified ones. The developed algorithm achieved a true positive detection rate of 98%, 97%, 88%, and 95% for four duration-related dIBI groups that we subsequently defined. We benchmarked our algorithm with an expert diagnostic interpretation of EEG periods (1h long) and demonstrated its effectiveness in clinical practice. We show that the detection algorithm effectively discriminates challenging cases encountered within mild and moderate background abnormalities. The dIBI detection algorithm improves identification of neonates with good clinical outcome as compared to the classification based on the classical burst-suppression interburst interval.

Automated quantitative 3D analysis of aorta size, morphology, and mural calcification distributions.

The purpose of this work is to develop a fully automated pipeline to compute aorta morphology and calcification measures in large cohorts of CT scans that can be used to investigate the potential of these measures as imaging biomarkers of cardiovascular disease. The first step of the automated pipeline is aorta segmentation. The algorithm the authors propose first detects an initial aorta boundary by exploiting cross-sectional circularity of aorta in axial slices and aortic arch in reformatted oblique slices. This boundary is then refined by a 3D level-set segmentation that evolves the boundary to the location of nearby edges. The authors then detect the aortic calcifications with thresholding and filter out the false positive regions due to nearby high intensity structures based on their anatomical location. The authors extract the centerline and oblique cross sections of the segmented aortas and compute the aorta morphology and calcification measures of the first 2500 subjects from COPDGene study. These measures include volume and number of calcified plaques and measures of vessel morphology such as average cross-sectional area, tortuosity, and arch width. The authors computed the agreement between the algorithm and expert segmentations on 45 CT scans and obtained a closest point mean error of 0.62 ± 0.09 mm and a Dice coefficient of 0.92 ± 0.01. The calcification detection algorithm resulted in an improved true positive detection rate of 0.96 compared to previous work. The measurements of aorta size agreed with the measurements reported in previous work. The initial results showed associations of aorta morphology with calcification and with aging. These results may indicate aorta stiffening and unwrapping with calcification and aging. The authors have developed an objective tool to assess aorta morphology and aortic calcium plaques on CT scans that may be used to provide information about the presence of cardiovascular disease and its clinical impact in smokers.

Low-power adaptive spike detector based on a sigma-delta control loop.

This paper presents a resources-optimized digital action potential (AP) detector featuring an adaptive threshold based on a new Sigma-delta control loop. The proposed AP detector is optimized for utilizing low hardware resources, which makes it suitable for implementation on most popular low-power microcontrollers units (MCU). The adaptive threshold is calculated using a digital control loop based on a Sigma-delta modulator that precisely estimates the standard deviation of the amplitude of the neuronal signal. The detector was implemented on a popular low-power MCU and fully characterized experimentally using previously recorded neural signals with different signal-to-noise ratios. A comparison of the obtained results with other thresholding approaches shows that the proposed method can compete with high performance and highly resources demanding spike detection approaches while achieving up to 100% of true positive detection rate at high SNR, and up to 63% for an SNR as low as 0 dB, while necessitating an execution time as low as 11 μs with the MCU operating at 8 MHz.

Multichannel spike detector with an adaptive threshold based on a Sigma-delta control loop.

In this paper, we present a digital spike detector using an adaptive threshold which is suitable for real time processing of 32 electrophysiological channels in parallel. Such a new scheme is based on a Sigma-delta control loop that precisely estimates the standard deviation of the amplitude of the noise of the input signal to optimize the detection rate. Additionally, it is not dependent on the amplitude of the input signal thanks to a robust algorithm. The spike detector is implemented inside a Spartan-6 FPGA using low resources, only FPGA basic logic blocks, and is using a low clock frequency under 6 MHz for minimal power consumption. We present a comparison showing that the proposed system can compete with a dedicated off-line spike detection software. The whole system achieves up to 100% of true positive detection rate for SNRs down to 5 dB while achieving 62.3% of true positive detection rate for an SNR as low as -2 dB at a 150 AP/s firing rate.

XML-AD: Detecting anomalous patterns in XML documents

Many information systems use XML documents to store data and to interact with other systems. Abnormal documents, which can be the result of either an on-going cyber attack or the actions of a benign user, can potentially harm the interacting systems and are therefore regarded as a threat. In this paper we address the problem of anomaly detection and localization in XML documents using machine learning techniques. We present XML-AD – a new XML anomaly detection framework. Within this framework, an automatic method for extraction of feature from XML documents as well as a practical method for transforming XML features into vectors of fixed dimensionality was developed. With these two methods in place, the XML-AD framework makes it possible to utilize general learning algorithms for anomaly detection. The core of the framework consists of a novel multi-univariate anomaly detection algorithm, ADIFA. The framework was evaluated using four XML documents datasets which were obtained from real information systems. It achieved over 89% true positive detection rate with less than 0.2% of false positives.

Video-based vehicle detection method

This paper considers the vehicle video detection and tracking problem. We propose a method based on the partitioning a video into blocks of equal length. Vehicles are detected in the first and last frames of each block. After that, locations of the object in the intermediate frames can be recovered. Experimental results show above 70% true positive detection rate; the average false positives per frame is 0.3.

A machine learning approach to crater detection from topographic data

Craters are distinctive features on the surfaces of most terrestrial planets. Craters reveal the relative ages of surface units and provide information on surface geology. Extracting craters is one of the fundamental tasks in planetary research. Although many automated crater detection algorithms have been developed to exact craters from image or topographic data, most of them are applicable only in particular regions, and only a few can be widely used, especially in complex surface settings. In this study, we present a machine learning approach to crater detection from topographic data. This approach includes two steps: detecting square regions which contain one crater with the use of a boosting algorithm and delineating the rims of the crater in each square region by local terrain analysis and circular Hough transform. A new variant of Haar-like features (scaled Haar-like features) is proposed and combined with traditional Haar-like features and local binary pattern features to enhance the performance of the classifier. Experimental results with the use of Mars topographic data demonstrate that the developed approach can significantly decrease the false positive detection rate while maintaining a relatively high true positive detection rate even in challenging sites.

Assessing environmental DNA detection in controlled lentic systems.

Little consideration has been given to environmental DNA (eDNA) sampling strategies for rare species. The certainty of species detection relies on understanding false positive and false negative error rates. We used artificial ponds together with logistic regression models to assess the detection of African jewelfish eDNA at varying fish densities (0, 0.32, 1.75, and 5.25 fish/m3). Our objectives were to determine the most effective water stratum for eDNA detection, estimate true and false positive eDNA detection rates, and assess the number of water samples necessary to minimize the risk of false negatives. There were 28 eDNA detections in 324, 1-L, water samples collected from four experimental ponds. The best-approximating model indicated that the per-L-sample probability of eDNA detection was 4.86 times more likely for every 2.53 fish/m3 (1 SD) increase in fish density and 1.67 times less likely for every 1.02 C (1 SD) increase in water temperature. The best section of the water column to detect eDNA was the surface and to a lesser extent the bottom. Although no false positives were detected, the estimated likely number of false positives in samples from ponds that contained fish averaged 3.62. At high densities of African jewelfish, 3–5 L of water provided a >95% probability for the presence/absence of its eDNA. Conversely, at moderate and low densities, the number of water samples necessary to achieve a >95% probability of eDNA detection approximated 42–73 and >100 L, respectively. Potential biases associated with incomplete detection of eDNA could be alleviated via formal estimation of eDNA detection probabilities under an occupancy modeling framework; alternatively, the filtration of hundreds of liters of water may be required to achieve a high (e.g., 95%) level of certainty that African jewelfish eDNA will be detected at low densities (i.e., <0.32 fish/m3 or 1.75 g/m3).

Current automated 3D cell detection methods are not a suitable replacement for manual stereologic cell counting

Stereologic cell counting has had a major impact on the field of neuroscience. A major bottleneck in stereologic cell counting is that the user must manually decide whether or not each cell is counted according to three-dimensional (3D) stereologic counting rules by visual inspection within hundreds of microscopic fields-of-view per investigated brain or brain region. Reliance on visual inspection forces stereologic cell counting to be very labor-intensive and time-consuming, and is the main reason why biased, non-stereologic two-dimensional (2D) “cell counting” approaches have remained in widespread use. We present an evaluation of the performance of modern automated cell detection and segmentation algorithms as a potential alternative to the manual approach in stereologic cell counting. The image data used in this study were 3D microscopic images of thick brain tissue sections prepared with a variety of commonly used nuclear and cytoplasmic stains. The evaluation compared the numbers and locations of cells identified unambiguously and counted exhaustively by an expert observer with those found by three automated 3D cell detection algorithms: nuclei segmentation from the FARSIGHT toolkit, nuclei segmentation by 3D multiple level set methods, and the 3D object counter plug-in for ImageJ. Of these methods, FARSIGHT performed best, with true-positive detection rates between 38 and 99% and false-positive rates from 3.6 to 82%. The results demonstrate that the current automated methods suffer from lower detection rates and higher false-positive rates than are acceptable for obtaining valid estimates of cell numbers. Thus, at present, stereologic cell counting with manual decision for object inclusion according to unbiased stereologic counting rules remains the only adequate method for unbiased cell quantification in histologic tissue sections.

Low-power high-sensitivity spike detectors for implantable VLSI neural recording microsystems

A spike detector has become a necessity of a contemporary multichannel neural recording microsystem for data-compression. This paper proposes two spike detection algorithms, frequency-enhanced nonlinear energy operator (fNEO) and energy-of-derivative (ED), to solve the sensitivity degradation suffered by the conventional nonlinear energy operator (NEO) at the presence of large-amplitude baseline interferences. The efficiency of NEO, fNEO and ED algorithms are evaluated with Simulink programs firstly and then implemented into three low-power spike detectors with a standard 0.13- $$\mu m$$ μ m CMOS process. To achieve a low-power design, subthreshold CMOS analog multipliers, derivatives and adders are developed to work with a low supply voltage, 0.5 V. The power dissipation of the proposed fNEO spike detector and ED spike detector are only 258.7 and 129.4 nW, respectively. The quantitative investigation shown in the paper indicates that both fNEO and ED spike detectors achieves superior performance than the conventional NEO spike detector. Considering its lowest power dissipation, the ED spike detector is selected for our application. Further statistical evaluations based on the true positive and false positive detection rate proves that the ED spike detectors achieves higher detection rate than that of the conventional NEO spike detector but dissipates 48 % less power.

Local tumour progression after loco-regional therapy of hepatocellular carcinomas: Value of fusion imaging-guided radiofrequency ablation

To assess the value of fusion imaging of real-time ultrasonography (US) with liver computed tomography (CT)/magnetic resonance imaging (MRI) images for planning US of radiofrequency ablation (RFA) in improving conspicuity of the lesions and reducing false-positive detection of local tumour progression (LTP) found after transcatheter arterial chemoembolization (TACE) or RFA of hepatocellular carcinoma (HCC). This study was approved by the institutional review board and informed consent was waived. Fifty patients with LTP (mean ± SD, 1.5 ± 0.6 cm; range 0.5-3 cm) detected at follow-up CT or MRI were included. Planning US was performed by two radiologists using conventional US first and fusion imaging later in the same session. False-positive detection rates were assessed using conventional US based on the results of fusion imaging. The number cases of initially invisible tumours on conventional US that became visible after image fusion were also evaluated. The true-positive detection rate and conspicuity scores of the index tumours were compared between conventional US and fusion imaging. On conventional US, 40 (80%) out of 50 HCCs with LTP were identified. However, the false-positive detection rate of conventional US was 12.5% (5/40). Out of 10 initially invisible HCCs with LTP on conventional US, six (60%) became visible after image fusion. The true-positive detection rate on conventional US was 70% (35/50), whereas it was increased to 92% (46/50) after image fusion (p = 0.0026). Fusion imaging can improve the conspicuity of lesions and reduce the false-positive detection of LTP after TACE or RFA.

Application of Receiver Operating Characteristic (ROC) Curves for Explosives Detection Using Different Sampling and Detection Techniques

Reported for the first time are receiver operating characteristic (ROC) curves constructed to describe the performance of a sorbent-coated disk, planar solid phase microextraction (PSPME) unit for non-contact sampling of a variety of volatiles. The PSPME is coupled to ion mobility spectrometers (IMSs) for the detection of volatile chemical markers associated with the presence of smokeless powders, model systems of explosives containing diphenylamine (DPA), 2,4-dinitrotoluene (2,4-DNT) and nitroglycerin (NG) as the target analytes. The performance of the PSPME-IMS was compared with the widely accepted solid-phase microextraction (SPME), coupled to a GC-MS. A set of optimized sampling conditions for different volume containers (1–45 L) with various sample amounts of explosives, were studied in replicates (n = 30) to determine the true positive rates (TPR) and false positive detection rates (FPR) for the different scenarios. These studies were obtained in order to construct the ROC curves for two IMS instruments (a bench-top and field-portable system) and a bench top GC-MS system in low and high clutter environments. Both static and dynamic PSPME sampling were studied in which 10–500 mg quantities of smokeless powders were detected within 10 min of static sampling and 1 min of dynamic sampling.

Fusion Imaging–Guided Percutaneous Biopsy of Focal Hepatic Lesions With Poor Conspicuity on Conventional Sonography

The purpose of this study was to evaluate the effectiveness of real-time fusion imaging (sonography combined with computed tomography or magnetic resonance imaging) for percutaneous sonographically guided biopsy of focal hepatic lesions with poor sonographic conspicuity. This study was conducted as a retrospective analysis of a prospective database and was approved by the Institutional Review Board. Patients who had target lesions with poor conspicuity on B-mode sonography for percutaneous biopsy were enrolled in the study. Lesion conspicuity was assessed by conventional B-mode sonography first and then by fusion imaging later in the same session. We compared lesion conspicuity and detection rates between B-mode sonography and fusion imaging and evaluated how many cases of initially invisible lesions on B-mode sonography became visible on fusion imaging. The technical success rate was evaluated on the basis of the final diagnoses, which were established by pathologic examination of the biopsy specimens as well as follow-up clinical and radiologic examinations. A total of 22 patients were enrolled in the study. On fusion imaging, lesion conspicuity was increased in 63.6% of focal hepatic lesions (14 of 22). Moreover, 66.7% of lesions (6 of 9) that were invisible on B-mode sonography became visible on fusion imaging. The true-positive detection rate was significantly different between B-mode sonography and fusion imaging (9 of 22 versus 19 of 22; P = .0044). Percutaneous biopsy was performed for all lesions, including 3 target lesions that were invisible even on fusion imaging. The technical success rate was 95.5% (21 of 22). Fusion imaging is effective for percutaneous biopsy of focal hepatic lesions with poor sonographic conspicuity.

High spatial resolution mapping of elevated atmospheric carbon dioxide using airborne imaging spectroscopy: Radiative transfer modeling and power plant plume detection

Carbon dioxide is emitted from the combustion of fossil fuels and is an important contributor to anthropogenic climate change. Multiple current and planned satellite missions are designed to quantify atmospheric carbon dioxide concentrations on a global scale, but most of these sensors do not have the spatial resolution necessary to resolve point sources such as fossil fuel power plants. Airborne imaging spectrometer data, such as those from the Airborne Visible InfraRed Imaging Spectrometer (AVIRIS), can have multiple, contiguous bands covering shortwave infrared (SWIR) absorption features produced by carbon dioxide. Therefore, high spatial resolution data from AVIRIS-like sensors may offer a means for detecting plumes and retrieving carbon dioxide concentrations for point source emissions. The objectives of this study include modeling minimum carbon dioxide anomalies detectable in AVIRIS data under different conditions and applying a Cluster-Tuned Matched Filter for detection of carbon dioxide plumes in simulated data and in AVIRIS images acquired over power plants. Radiative transfer simulations were used to model the residual radiance produced by increased absorption by carbon dioxide as concentration was elevated above background levels within a 0–500m layer. Carbon dioxide anomalies, surface reflectance, water vapor concentration, solar zenith angle, sensor height, and aerosol scattering were varied in simulation sets and the resulting residual radiance spectra were compared against noise equivalent delta radiance (NEdL) for the “classic” and “next generation” AVIRIS instruments. Sensitivity to carbon dioxide anomalies improved with increased surface reflectance and declined with increased water vapor concentration, solar zenith angle, sensor height, and aerosol scattering. Zero to 500m concentration anomalies as low as 100 parts per milion by volume (ppm) for AVIRIS C and 25ppm for AVIRIS NG produced residual radiance values that exceeded SWIR NEdL. Carbon dioxide concentrations modeled for a generic power plant emissions scenario using a plume dispersion model were combined with randomly-generated reflectance spectra to create simulated images with varying surface reflectance and NEdL. For these simulated images, true positive and false positive detection rates improved as background reflectance increased and as NEdL decreased. Apparent plumes were detected in all four AVIRIS C images acquired over power plants, although the characteristics of the plumes varied according to solar-plume-sensor geometry. Improvements in modeling may allow retrieval of plume concentration, providing a valuable means for quantifying point source emissions and a basis for comparison with column concentrations retrieved from in situ measurements and coarse resolution satellite data.

Planning US for Percutaneous Radiofrequency Ablation of Small Hepatocellular Carcinomas (1–3 cm): Value of Fusion Imaging with Conventional US and CT/MR Images

PurposeTo assess whether fusion of conventional ultrasonography (US) with liver computed tomography/magnetic resonance images for planning US for percutaneous radiofrequency (RF) ablation can reduce false-positive detection and enhance lesion detectability of small hepatocellular carcinomas (HCCs) on conventional US. Materials and MethodsThis retrospective study was approved by the institutional review board, and informed consent was waived. A total of 137 patients with single HCCs (mean ± standard deviation, 1.7 ± 0.6 cm; range, 1.1–3.0 cm) were included. Planning US was performed by two radiologists by using conventional US first and fusion imaging later in the same session. The false-positive detection rate of conventional US was assessed with the results of fusion imaging used as a reference standard. True-positive detection rates on conventional US and fusion imaging were compared by McNemar test. Initially undetectable HCCs on conventional US that became detectable after image fusion were also assessed. ResultsThe false-positive detection rate of conventional US was 7.7% (nine of 117). Overall true-positive detection rates on conventional US and fusion imaging were 78.8% (108 of 137) and 90.5% (124 of 137), respectively (P = .0002); the rates were significantly different between conventional US and fusion imaging for HCCs smaller than 2.0 cm, but not for HCCs 2.0 cm or larger. Of 20 initially undetectable HCCs on conventional US, nine (45.0%) became detectable after image fusion. ConclusionsFusion imaging for planning US for percutaneous RF ablation can reduce false-positive detection and enhance lesion detectability of small HCCs on conventional US.

Robust pixel-based classification of obstacles for robotic harvesting of sweet-pepper

Sweet-pepper plant parts should be distinguished to construct an obstacle map to plan collision-free motion for a harvesting manipulator. Objectives were to segment vegetation from the background; to segment non-vegetation objects; to construct a classifier robust to variation among scenes; and to classify vegetation primarily into soft (top of a leaf, bottom of leaf and petiole) and hard obstacles (stem and fruit) and secondarily into five plant parts: stem, top of a leaf, bottom of a leaf, fruit and petiole. A multi-spectral system with artificial lighting was developed to mitigate disturbances caused by natural lighting conditions. The background was successfully segmented from vegetation using a threshold in a near-infrared wavelength (>900nm). Non-vegetation objects occurring in the scene, including drippers, pots, sticks, construction elements and support wires, were removed using a threshold in the blue wavelength (447nm). Vegetation was classified, using a Classification and Regression Trees (CART) classifier trained with 46 pixel-based features. The Normalized Difference Index features were the strongest as selected by a Sequential Floating Forward Selection algorithm. A new robust-and-balanced accuracy performance measure PRob was introduced for CART pruning and feature selection. Use of PRob rendered the classifier more robust to variation among scenes because standard deviation among scenes reduced 59% for hard obstacles and 43% for soft obstacles compared with balanced accuracy. Two approaches were derived to classify vegetation: Approach A was based on hard vs. soft obstacle classification and Approach B was based on separability of classes. Approach A (PRob=58.9) performed slightly better than Approach B (PRob=56.1). For Approach A, mean true-positive detection rate (standard deviation) among scenes was 59.2 (7.1)% for hard obstacles, 91.5 (4.0)% for soft obstacles, 40.0 (12.4)% for stems, 78.7 (16.0)% for top of a leaf, 68.5 (11.4)% for bottom of a leaf, 54.5 (9.9)% for fruit and 49.5 (13.6)% for petiole. These results are insufficient to construct an accurate obstacle map and suggestions for improvements are described. Nevertheless, this is the first study that reports quantitative performance for classification of several plant parts under varying lighting conditions.

Pixel classification and post-processing of plant parts using multi-spectral images of sweet-pepper

As part of the development of a sweet-pepper harvesting robot, obstacles should be detected. Objectives were to classify sweet-pepper vegetation into five plant parts: stem, top of a leaf (TL), bottom of a leaf (BL), fruit and petiole (Pet); and to improve classification results by post-processing. A multi-spectral imaging set-up with artificial lighting was developed to acquire images of sweet-pepper plants. The background was segmented from the vegetation and vegetation was classified into five plant parts, through a sequence of four two-class classification problems. True-positive detection rate/scaled false-positive rate achieved, on a pixel basis, were 40.0/179% for stem, 78.7/59.2% for top of a leaf (TL), 68.5/54.8% for bottom of a leaf (BL), 54.5/17.2% for fruit and 49.5/176.0% for petiole (Pet), before post-processing. The opening operations applied were unable to remove false stem detections to an acceptable rate. Also, many false detections of TL (>10%), BL (14%) and Pet (>15%) remained after post-processing, but these false detections are not critical for the application because these three plant parts are soft obstacles. Furthermore, results indicate that TL and BL can be distinghuished. Green fruits were post-processed using a sequence of fill-up, opening and area-based segmentation. Several area-based thresholds were tested and the most effective threshold resulted in a true-positive detection rate, on a blob basis, of 56.7 % and a scaled false-positive detection rate of 6.7 % for green fruits (N=60). Such fruit detection rates are a reasonable starting point to detect obstacles for sweet-pepper harvesting. But, additional work is required to complement the obstacle map into a complete representation of the environment.