No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Classification of weed species using artificial neural networks based on color leaf texture feature
Abstract
The potential impact of herbicide utilization compel people to use new method of weed control. Selective herbicide application
is optimal method to reduce herbicide usage while maintain weed control. The key of selective herbicide is how to discriminate
weed exactly. The HIS color co-occurrence method (CCM) texture analysis techniques was used to extract four texture parameters:
Angular second moment (ASM), Entropy(E), Inertia quadrature (IQ), and Inverse difference moment or local homogeneity (IDM).The
weed species selected for studying were Arthraxon hispidus, Digitaria sanguinalis, Petunia, Cyperus, Alternanthera Philoxeroides
and Corchoropsis psilocarpa. The software of neuroshell2 was used for designing the structure of the neural network, training
and test the data. It was found that the 8-40-1 artificial neural network provided the best classification performance and
was capable of classification accuracies of 78%.
... Its main advantages are the fact that it is non-parametric, it enables nonlinear solutions, and considers several explanatory variables simultaneously (Niska et al., 2010). This technique has been successfully used in the identification of Banksia integrifólia genotypes (Pandolfi et al., 2009), Camellia species (Lu et al., 2012), weed species (Li et al., 2009) and wheat plants among weed herbs (Gomez-Casero et al., 2010). ...
... This accuracy rate decreased with the reduction of the number of neurons in the hidden layer. When only one neuron was used in the hidden layer, in the 1000 trainings, there was a mean accuracy rate of 78 %. Li et al. (2009), studying weeds classification, also observed a decrease in the efficiency of neural networks with a reduced number of neurons in the hidden layer. Figure 2 shows the network architecture with six neurons in the hidden layer, which was more efficient (Figure 1). ...
... Figure 2 shows the network architecture with six neurons in the hidden layer, which was more efficient (Figure 1). In a similar study, Li et al. (2009) used 60 neurons in the hidden layer, and found a 78 % accuracy rate. The greater efficiency found in the present study (100 %) may be explained by the greater number of plants evaluated (564 plants), since for more efficient network training, it is important the availability of large data sets (Azevedo et al., 2015). ...
Germplasm classification by species requires specific knowledge on/of the culture of interest. Therefore, efforts aimed at automation of this process are necessary for the efficient management of collections. Automation of germplasm classification through artificial neural networks may be a viable and less laborious strategy. The aims of this study were to verify the classification potential of Capsicum accessions regarding/ the species based on morphological descriptors and artificial neural networks, and to establish the most important descriptors and the best network architecture for this purpose. Five hundred and sixty-four plants from 47 Brazilian Capsicum accessions were evaluated. Neural networks of multilayer perceptron type were used in order to automate the species identification through 17 morphological descriptors. Six network architectures were evaluated, and the number of neurons in the hidden layer ranged from 1 to 6. The relative importance of morphological descriptors in the classification process was established by Garson's method. Corolla color, corolla spot color, calyx annular constriction, fruit shape at pedicel attachment, and fruit color at mature stage were the most important descriptors. The network architecture with 6 neurons in the hidden layer is the most appropriate in this study. The possibility of classifying Capsicum plants regarding/ the species through artificial neural networks with 100 % accuracy was verified.
... Real field images were used in this computer vision-based automated weed detection system. In [13], crops and weeds were discriminated using texture features extracted using a color co-occurrence matrix (CCM) and artificial neural networks (ANN). In [14], weeds were recognized based on shape features. ...
Weed management is one of the important tasks in agriculture. Weeds in rice fields are usually managed using three ways - chemical herbicides, mechanical weeders, and manual weeding. Manual weeding becomes a problem when there is a shortage of agricultural laborers. Mechanical weeders are not suitable for direct-seeded rice fields. Chemical herbicides are not advisable especially when farmers do not know about site-specific weed management. Site-specific weed management is using the right herbicide in the right amount. Therefore, this paper investigates computer vision-based deep learning techniques with transfer learning classifying three types of weeds in paddy fields, namely sedges, grasses, and broadleaved weeds so that the right herbicide is recommended to the farmers. This would reduce the broadcast application and the overuse of the herbicides, thereby limiting the negative impact of the chemical herbicides on the environment. This research work shows promising results with an accuracy around 90% and thus encourages the development of digital agriculture.
... The identification of agricultural products using ANNs is done by learning from data created experimentally (Li et al., 2009). A common category of ANN is the backpropagation neural network, which is used to group features into different classes by identifying common appearances among the samples of the known feature class. ...
This study aimed to develop a method for identifying different cultivars of Indian jujube fruits (Ziziphus mauritiana Lamk.) based on a single Indian jujube fruit color and morphological attributes using an artificial neural network (ANN) classifier. Eleven Indian jujube fruit cultivars were collected during winter of season 2020 from a local orchard located at Riyadh region, Saudi Arabia to measure their lengths, major diameters, and minor diameters. Different morphological descriptors were calculated, including the arithmetic mean diameter, the sphericity percent, and the surface area. Moreover, the color values of L*, a*, and b* of the skin of fruits were recorded. The ANN classifier was used to identify the appropriate class of Indian jujube fruit by using a combination of morphological and color descriptors. The proposed method achieved an overall identification rate of 98.39% and 97.56% in training and testing phases, respectively. In addition to color and morphological features, ANN classifier is a useful tool for identifying Indian jujube fruit cultivars and circumventing the difficulties met during fruit grading.
... Prediction via ANN is not like modeling and simulation, but by learning from data generated experimentally or using validated models. ANNs differ from conventional programs in their ability to learn about the system to be modeled without prior knowledge of the relationship of the process variables (Li et al., 2009). It is one of the data mining methods that are developed in various contexts such as function approximation, modeling, and classification of agricultural products (Marini et al, 2004). ...
One of the new crop varieties that have been adopted for high yield is the Egyptian faba bean. However, poor-quality faba bean has reduced economic value. Quality evaluation is thus important and can be performed using computational intelligence. We developed a robust method based on morphological features and artificial neural network for quality grading and classification of Egyptian faba-bean seeds, covering five varieties: Giza3, Giza461, Misr1, Nobarya1, and Sakha1. Fifteen seed morphological features were then calculated, and artificial neural networks classified faba beans into different varieties. The results indicated an overall classification accuracy of 77.5% was achieved in training phase and it was 100% when testing dataset was used. The preliminary work presented in this paper could be further enhanced by real time faba beans identification by capturing seed morphological features through the help of digital images.
... The feature selection process ensures the elimination of irrelevant features, which would otherwise compromise the ability of the machine-learning models [25]. Artificial neural networks (ANNs), one of the most powerful and advanced machine-learning classifiers, have been frequently used for weed detection and mapping [26][27][28]. ANNs, in general, are the computing systems that mimic the biological neural networks, comprising of three main systems namely the input layer to receive the data, hidden layer(s) to learn the pattern in the data, and output layer to provide the best parameters for classification [29]. ...
Italian ryegrass (Lolium perenne ssp. multiflorum (Lam) Husnot) is a troublesome weed species in wheat (Triticum aestivum) production in the United States, severely affecting grain yields. Spatial mapping of ryegrass infestation in wheat fields and early prediction of its impact on yield can assist management decision making. In this study, unmanned aerial systems (UAS)-based red, green and blue (RGB) imageries acquired at an early wheat growth stage in two different experimental sites were used for developing predictive models. Deep neural networks (DNNs) coupled with an extensive feature selection method were used to detect ryegrass in wheat and estimate ryegrass canopy coverage. Predictive models were developed by regressing early-season ryegrass canopy coverage (%) with end-of-season (at wheat maturity) biomass and seed yield of ryegrass, as well as biomass and grain yield reduction (%) of wheat. Italian ryegrass was detected with high accuracy (precision = 95.44 ± 4.27%, recall = 95.48 ± 5.05%, F-score = 95.56 ± 4.11%) using the best model which included four features: hue, saturation, excess green index, and visible atmospheric resistant index. End-of-season ryegrass biomass was predicted with high accuracy (R 2 = 0.87), whereas the other variables had moderate to high accuracy levels (R 2 values of 0.74 for ryegrass seed yield, 0.73 for wheat biomass reduction, and 0.69 for wheat grain yield reduction). The methodology demonstrated in the current study shows great potential for mapping and quantifying ryegrass infestation and predicting its competitive response in wheat, allowing for timely management decisions.
... In literature, only a few research works have been carried out which concentrates on only color features for crop and weed discrimination. In [21], weed and crop discrimination are done using the color feature. HIS based Color co-occurrence matrix(CCM) is used to extract Entropy(E), Inertia quadrature (IQ), Inverse difference moment or local homogeneity (IDM) and Angular second moment (ASM). ...
Weed detection in paddy fields using robotic vision is still a challenging task. The main reason for this being lack of dataset. In this research work, the creation of paddy crop and weeds image dataset has been described. The images were acquired using a digital camera under natural lighting conditions. For every image in the dataset, its annotated image was created using manual annotation. Annotation was carried out after plant segmentation from the soil background. Dataset consists of 300 images. In addition, classification of paddy crop and two types of weeds namely, sedges and grass-type weeds has been done using only color features. Usually, crop and weed discrimination is done using texture, shape and color features. However, using all three features may result in a computationally intensive system. Color features are extracted using a novel approach based on Speeded-up Robust Features (SURF). Random Forest, K-Nearest Neighbors (K-NN) and Least Squares Support Vector Machine (LSSVM) classifiers have been used for the classification of the paddy crop and two types of weeds. An accuracy around 86% was obtained by all the classifiers. This indicates color features can be relied upon in discrimination between paddy crop and sedges and grass-type weeds.
... Kahraman et al. utilized Gabor filters for the retrieval of vehicle lice plate textures [21]. Li et al. used color co-occurrence matrices to represent the texture features of four leaf types [22]. ...
Several research studies have been recently initiated to investigate the use of construction site images for automated infrastructure inspection, progress monitoring, etc. In these studies, it is always necessary to extract material regions (concrete or steel) from the images. Existing methods made use of material's special color/texture ranges for material information retrieval, but they do not sufficiently discuss how to find these appropriate color/texture ranges. As a result, users have to define appropriate ones by themselves, which is difficult for those who do not have enough image processing background. This paper presents a novel method of identifying concrete material regions using machine learning techniques. Under the method, each construction site image is first divided into regions through image segmentation. Then, the visual features of each region are calculated and classified with a pre-trained classifier. The output value determines whether the region is composed of concrete or not. The method was implemented using C++ and tested over hundreds of construction site images. The results were compared with the manual classification ones to indicate the method's validity.
... In the bibliography we can find some approaches of robotic systems used in the agriculture to perform different maintenance tasks, [1,4,3,10,9]. In general, most of the approaches are based on the use of a single complex robot which performs specific actions on the environment, but we have not found any cooperative system, with some different robots, for agricultural tasks. ...
In agricultural robotics, as in other robotic systems, one of the most important parts is the control architecture. This paper describes the definition of a new control architecture specially designed for groups of robots in charge of doing maintenance tasks in agricultural environments. This architecture has been developed having in mind principles as scalability, code reuse, abstraction hardware and data distribution. Moreover, it is important that the control architecture can allow coordination and cooperation among the different elements in the system. The architecture presented in this paper implements all these concepts by means of the integration of different systems, such as Player, JADE and HLA. The most important system is HLA because it not only allows the data distribution and implicit communication among the parts of the system, but also allows to operate with simulated and real entities at the same time, allowing the use of hybrid systems in the development of applications.
Se presenta el desarrollo de un sistema capaz de estimar la materia orgánica, hierro y capacidad de intercambio catiónico del suelo, a través del uso de Redes Neuronales Artificiales (RNA) y Análisis de Componentes Principales (ACP). Las variables empleadas para realizar análisis fueron: textura, histograma de color HSL y RGB de las imágenes, las cuales fueron correlacionadas con las propiedades obtenidas en los análisis de laboratorio convencionales. ABSTRACT The development of a system capable of estimating the organic matter, iron and capacity for cationic exchange of the soil, through the use of Artificial Neural Network (ANN) and principal component analysis (PCA), is presented. The variables used to perform the analysis were texture, HSL color histogram and RGB of the images, which were correlated with the properties obtained in conventional lab analyses.
From the last few years, more attention has been directed towards the usage of information technology in agriculture. This new way of farming offers the promise of improving farm profitability. Using Internet the farmers can collect data like geographical- referred yield, weather, soil and other important data related to farming. The aim is to use these data to produce area-specific crop production decisions. For increasing the production quality of crop soil plays very important role. To help the farmer in deciding how to increase the crop quality based on soil. We have proposed, soil image recognizer and classifier, which classifies soil image samples based on their color and morphological features. Different types of soil image samples considered like red soil, black soil, black cotton soil. Using color and morphological features a Neural Network Based Classifier is designed. The effect of water on the soil image classifier is analyzed by adding the water into different portions of soil samples. The accuracy of the soil image classifier is improved by considering wet soil samples.
The control architecture is one of the most important part of agricultural robotics and other robotic systems. Furthermore its importance increases when the system involves a group of heterogeneous robots that should cooperate to achieve a global goal. A new control architecture is introduced in this paper for groups of robots in charge of doing maintenance tasks in agricultural environments. Some important features such as scalability, code reuse, hardware abstraction and data distribution have been considered in the design of the new architecture. Furthermore, coordination and cooperation among the different elements in the system is allowed in the proposed control system. By integrating a network oriented device server Player, Java Agent Development Framework (JADE) and High Level Architecture (HLA), the previous concepts have been considered in the new architecture presented in this paper. HLA can be considered the most important part because it not only allows the data distribution and implicit communication among the parts of the system but also allows to simultaneously operate with simulated and real entities, thus allowing the use of hybrid systems in the development of applications.
Color co-occurrence method (CCM) texture statistics were used as input variables for a backpropagation (BP) neural network weed classification model. Thirty-three unique CCM texture statistic inputs were generated for 40 images per class, within a six class data set. The following six classes were studied: giant foxtail, large crabgrass, common lambsquarter, velvetleaf, ivyleaf morningglory, and clear soil surface. The texture data was used to build six different input variable models for the BP network, consisting of various combinations of hue, saturation, and intensity (HSI) color texture statistics. The study evaluated classification accuracy as a function of network topology, and training parameter selection. In addition, training cycle requirements and training repeatability were studied. The BP topology evaluation consisted of a series of tests on symmetrical two hidden-layer network, a test of constant complexity topologies, and tapered topology networks. The best symmetrical BP network achieved a 94.7% classification accuracy for a model consisting of 11 inputs, five nodes at each of the two hidden layers and six output nodes (11 x 5 x 5 x 6 BP network). A tapered topology (11 x 12 x 6 x 6 BP network) out performed all other BP topologies with an overall accuracy of 96.7% and individual class accuracies of 90.0% or higher.
A novel texture-based weed classification method was developed. The method comprised a low-level Gabor wavelets-based feature extraction algorithm and a high-level neural network-based pattern recognition algorithm. The design strategy simulated the function of the human visual system, which uses low-level receptors for earl y stage vision processing and high-level cognition for pattern recognition and image understanding . This model was specifically developed to classify images into broadleaf and grass categories for real-time selective herbicide application. The results showed that the method is capable of performing texture -based broadleaf and grass classification effectively and with 100 percent classification accuracy over 40 sample images with 20 samples from each class. Based on the elapsed time to do weed classification, the method meets real-time constraints.
Take the weeds in wheat fields as the research object, a method of weed detection by using the texture and position features was studied. According to the position feature of drilled crops that were regularly sown as a constant row space, the pixel-histogram method was used to determine the central line and the width of crop row. As a result, weeds between crop rows were detected. Moreover, the block of texture was selected on the basis of the central line of crop row. The co-occurrence matrixes of the H color space that was quantified 8 levels were computed. Based on that, five texture parameters were extracted. Then, the K-means clustering method was used to recognize weeds within crop rows. The result of research showed that the correct classification of weeds was 93% and the mistake classification of crops was 7%.
Based on the experiment, a BP neural network model was developed to map the relationship of process variables and distributing quality coefficient of Profile Modeling Spray of the Fruit Trees. The structure and parameters of BP neural network were optimized by using the methods of orthogonal experiments, intercross evaluating and training, normalization of the network training set and principal component analysis. The result shows that the correlation coefficient R between simulating outputs of the BP network and the results of experiments is 0.99, which has wide adaptability. And the BP network can be used conveniently to carry out various quantificational calculations. For example, the distributing quality coefficient under specific process variables can be estimated, or appropriate spray variables can be determined according to given production target.
To improve the accuracy of the automatic detection and classification of the moldy peanut kernels by using the machine vision, a discriminating method was developed based on image features of the peanut kernels and artificial neural network. First, a segmented colorful image of peanut kernel was obtained by edges detecting, filtering, filling and composing and so on, to remove the background to those relatively complete edge-preserved B branch grey image with less noise. The image characteristics parameters such as the color parameters H, I, S, and veins characteristics parameters RW, GW, BW were used as the input to the neural network set up by MATLAB. Three outputs were defined as 100, 010, 001 which represented the normal level, slightly moldy level and severely moldy level, and an identifying model of the neural network was set up between the feature parameters and moldy grade of the peanut. The results of the experiment show that the accuracies of the identification of the method are 95% for normal peanut kernels, 90% for slightly moldy peanut kernels and 100% for severely moldy peanut kernels, respectively.
The environmental impact of herbicide utilization has stimulated research into new methods of weed control, such as selective herbicide application on highly infested crop areas. This research utilized the Color Co-occurrence Method (CCM) to determine whether traditional statistical discriminant analysis can be used to discriminate between six different classes of groundcover. The weed species evaluated were giant foxtail, crabgrass, common lambsquarter, velvetleaf, and ivyleaf morningglory, along with a soil image data set. The between species discriminant analysis showed that the CCM texture statistics procedure was able to classify between five weed species and soil with an accuracy of 93% using hue and saturation statistics, only. A significant accomplishment of this work was the elimination of the intensity texture features from the model, which reduces computational requirements by one-third.
A method of identifying plants based on color texture characterization of canopy sections was developed. Color co-occurrence matrices were derived from image matrices, one for each color attribute: intensity, saturation, and hue. Eleven texture features were calculated from each of the co-occurrence matrices. The 33 total color texture features were used in a discriminant analysis model to identify plants. Overall classification accuracy of 91% was achieved when this method was used to identify seven common cultivars of nursery stock. A total of 350 observations were used in the investigation. This method exhibited a significant improvement over previous methods which used intensity data only.
Advanced computer vision and statistical methods were employed for identifying living plants from soil/residue background for two species of grasses (Shattercane, Green Foxtail) and two broadleaf species (Velvetleaf, Red Root Pigweed) weeds. The excess green index method was used as a contrast enhancement for specifically identifying plant from soil regions. Excess green classified plant and soil regions correctly over the entire three-week observation period with high accuracies (99% plus). Plant and soil binary images were derived from excess green images and provided edge boundaries. These boundaries were used with corresponding gray scale images to extract four classical textural features for plants and soil: angular second moment, inertia, entropy, and local homogeneity. These features were derived from the co-occurrence matrix. Stepwise and canonical discriminant analyses were used to test the classification performance of the texture and excess green features. Discrimination models of local homogeneity, inertia, and angular second moment were found to classify grass and broadleaf categories of plants, with classification accuracies of 93 and 85%, respectively. Classification accuracies of individual species only ranged from 30 to 77%. Soil classification accuracies were also high for textural feature algorithms (97%). The time required to produce tokensets ranged from 15 to 20 s on a UNIX computer system. Additional time required for the system to reach a plant/soil classification ranged from 5 to 10 s. This translated into an overall system response time of 20 to 30 s, with the preprocessing step constituting the major part of the system response time.
Weeds are frequently distributed non-uniformly within fields. Herbicide use could be reduced either by applying it only to weed-infested areas or by applying a low dose rate to the whole field and normal dose rate to weed patches. A necessary prerequisite is an effective method of weed detection. Tractor-mounted detectors do not appear to be feasible. Thus real-time weed detection and sprayer control are not currently possible. It is shown, however, that spatially variable herbicide application using a sprayer location system and a field map of weed location has potential. The data map would be constructed from a number of weed-location techniques based on image analysis, such as tractor-mounted video cameras, aerial photographs and manual field observation.
Commercially available neural networks software was applied for prediction of processing parameters with reference to the product quality of the dehydrated cooked rice. These results were verified with experimental data. the experimental results indicate good concurrence with the predicted data. A small capacity vibrofluidized bed drier was used to conduct the experimental studies. the inlet air temperature of 160C to 180C, the inlet air velocity of 3 to 6 ms−1 and the resident time of 5 to 8 min were used in the study. the optimized process parameters were identified. the operation of the neural network model is discussed. the trained model can be applied in a nonlinear model predictive scheme to control the product moisture content.
The potential environmental impact of herbicide utilisation has stimulated research into new methods of weed control. Selective herbicide application offers the possibility to reduce herbicide usage while maintaining weed control. The research reported utilised colour co-occurrence method (CCM) texture analysis techniques to evaluate three different neural-network classifiers for potential use in real-time weed control systems. An image data set consisting of 33 unique texture features for each image in a six class data set (40 images per class) was generated for the following classes; foxtail, crabgrass, common lambsquarter, velvetleaf, morning-glory, and clear soil surface. The data sets were evaluated using a stepwise variable selection procedure to provide six unique data models separated according to hue-saturation-intensity colour features. A comparison study of the classification capabilities of three neural-network models (backpropagation, counterpropagation, and radial basis function) was conducted. It was found that the backpropagation neural-network classifier provided the best classification performance and was capable of classification accuracies of 97%, which exceeded traditional statistical classification procedure accuracy of 93%. When comparing the three neural-network methodologies, the backpropagation method not only achieved a higher classification accuracy, but also had less computational requirements.
Thesis (Ph. D.)--Ohio State University, 1986. Includes bibliographical references (leaves 171-177). Photocopy.
Machine vision system used for realtime detection inter-row weed. Transaction of the CSAE
- Mao Wenhua
- Wang Yiming
- Zhang Xiaochao
Mao Wenhua,Wang Yiming, Zhang Xiaochao et al. Machine vision system used for realtime detection inter-row weed. Transaction of the CSAE 2003,19(5),114-117.