Precision Farming
F. Nadernejad; D. M. Imani; M. R. Rasouli
Abstract
IntroductionSugarcane is a strategic agricultural product and increasing productivity and self-sufficiency in its production is of special importance. The most important product of sugarcane is sugar. Various factors like climatic and management conditions affect the yield of sugarcane and recoverable ...
Read More
IntroductionSugarcane is a strategic agricultural product and increasing productivity and self-sufficiency in its production is of special importance. The most important product of sugarcane is sugar. Various factors like climatic and management conditions affect the yield of sugarcane and recoverable sugar. Crop yield forecasting is one of the most important topics in precision agriculture, which is used to estimate yield, match product supply with demand and manage product to increase productivity. The purpose of this study is to predict and model the factors affecting sugar extracted from sugarcane (recoverable sugar) in the farms of Amir-Kabir sugarcane agro-industry Company of Khuzestan province using machine learning methods.Materials and MethodsTo conduct this study, data from the agro-industrial company Amir-Kabir in the province of Khuzestan from 2010 to 2017 were used. This data has 3223 records which include four sets of data: climate, soil, crop and farm management. This data includes continuous and discrete variables. Discrete variables include production management, soil type, farm, variety, age (cane class), the month of harvest and times irrigation. Continuous variables include area, chemical fertilizer consumption, water consumption per hectare, total water consumption, drain, crop season duration, yield (cane yield) soil EC, purity, time interval drying off to crop harvest, precipitation, min and max temperature, min and max relative humidity, wind speed and evaporation. The recoverable sugar variable is considered as the target variable and is divided into two classes, values greater than or equal to 9 are in the optimal class and less than 9 are in the undesirable class. The other variables are considered as predictor variables. For modeling using the Holdout method the data were randomly divided into two independent sets, a training set and a test set. 70% of the data which includes 2256 records were used for training and 30% of the data which includes 967 records were used for testing. The modeling of this study was performed with the Python programming language version 3.8.6 in the Jupyter notebook environment. Random Forest, Adaboost, XGBoost and SVM (support vector machine) algorithms were used for modeling.Results and DiscussionTo evaluate the models, metrics of accuracy, precision, recall, f1 score and k-fold cross validation were used. The XGBoost model with 94.8% accuracy on the training set and the Adaboost model with 92.4% accuracy on the test set, are the best models. Based on precision and recall metrics Adaboost model with 87% precision and SVM model with 87% recall have better performance than the other models. Based on Repeated 10-fold stratified cross validation using two repeats the SVM model with 92.3% accuracy is the best model. The variables of purity, time interval drying off to crop harvest and crop season duration are the most important variables in predicting the recoverable sugar.ConclusionIn this study a new approach based on machine learning methods for predicting recoverable sugar from sugarcane was presented. The most important innovation of this study is the simultaneous consideration of management and climatic factors, along with other factors such as soil and crop characteristics for modeling and classification the recoverable sugar percentage from sugarcane. The results show that the performance of all models is acceptable and machine learning methods and ensemble learning algorithms can be used to predict crop yield. The results of this study and the analysis of the rules obtained from the set of decision trees made in the random forest model can be used for managers of different agro-industries in determining appropriate strategies and preparing the conditions to achieve optimal production.For future research as well as policy making and decision making Amir-Kabir sugarcane agro-industry Company the following suggestions are offered: more samples can be used to obtain more reliable results. Also can be used Deep learning methods, time series analysis and image processing. Use of IOT equipment to collect and real-time processing data on Amir-Kabir sugarcane agro-industry farms.
Image Processing
Z. Azizpour; H. Vahedi; A. N. Lorestani
Abstract
IntroductionPistachio or Green Gold is one of the most important agricultural crops and is especially important for Iranian exports. A group of pistachio's pests mainly feed on pistachio, among which Idiocerus stali is very important. Conventional methods for identifying insects using identification ...
Read More
IntroductionPistachio or Green Gold is one of the most important agricultural crops and is especially important for Iranian exports. A group of pistachio's pests mainly feed on pistachio, among which Idiocerus stali is very important. Conventional methods for identifying insects using identification keys are time-consuming and costly. Due to the rapid development of the Pistachio industry, the use of artificial intelligence techniques such as image processing, for identification and population monitoring is highly recommended. On the other hand, little research was carried out on I. stali. Therefore, in this research, I. stali was selected as a target insect for the identification and counting on sticky yellow cards using image processing techniques and artificial neural networks. The purpose of this study was to determine the feasibility of I. stali identification algorithm by image processing, to determine the possibility of separation and counting of I. stali from other non-target insects by artificial neural network and to determine its accuracy in identification of I. Stali.Materials and MethodsIdiocerus stali was selected as the target insect for identification. Sticky yellow cards were used for collecting samples. Taking the photos with the help of a SONY Handycam Camera, which had a 12-megapixel resolution and G lens, was carried out (SONY, HDR-XR500, CMOS, SONY Lens G, Made in Japan). Then insects were counted on each card manually and the data was recorded. The data, which were digital images of yellow sticky cards, were imported into the MatLab R2017b software environment. A total of 357 color properties and 20 shape's features for the identification of I. stali were extracted by an image processing algorithm. Color properties were divided into two categories of mean and standard deviation and characteristics related to vegetation indices. An ANN-PSO (Artificial Neural Network hybrid method-Particle Swarm Optimization) algorithm was used to select the effective features. The selected effective characteristics for insect classification were: Color index for extra collective vegetation related to HSL color space, normalized difference index for LCH color space, gray channel for color space YCbCr, second component index minus third component for color space YCbCr, area and mean of the first, second and third components of color space Luv.Results and DiscussionComparing the results with the results of Qiao et al. (2008), we found that in his study, which divided the data into three categories, for medium and high-density groups, the detection rate was 95.2% and 94.6%, respectively. On the other hand, in low densities (less than 10 trapped insects); its detection rate was 72.9%, while the detection rate of the classifier system designed in this study for different densities of trapped insects, was identical and equal to 99.59%. Also, comparing the results of this study with Espinoza et al. (2016), we found that their algorithm in whiteflies detection had a high accuracy of about 0.96 on a sticky yellow card, while the Thrips identification algorithm accuracy was 0.92 on a sticky blue card. As stated above, the correct detection rate of I. stali by the algorithm designed in this study was 99.72%.ConclusionThe results showed the feasibility of the new method for identifying the pest insects without destroying them on the farm and in natural light conditions and in a short time and with very high accuracy. This suggests that this algorithm can be applied to the machine vision system and can be used in future in the construction of agricultural robots.
A. Moghimi; A. Sazgarnia; M. H. Aghkhani
Abstract
IntroductionPistachio production has been adversely affected by Psylla, which is a devastating insect. The primary goal of this study was to select sensitive spectral bands to distinguish pistachio leaves infected by Psylla from healthy leaves. Diagnosis of psylla disease before the onset of visual cues ...
Read More
IntroductionPistachio production has been adversely affected by Psylla, which is a devastating insect. The primary goal of this study was to select sensitive spectral bands to distinguish pistachio leaves infected by Psylla from healthy leaves. Diagnosis of psylla disease before the onset of visual cues is crucial for making decisions about topical garden management. Since it is not possible to diagnose psylla disease even after the onset of symptoms with the help of color images by drones, hyperspectral and multispectral sensors are needed. The main purpose of this study was to extract spectral bands suitable for distinguishing healthy leaves from psylla leaves. For this purpose, in this paper, a new method for selecting sensitive spectral properties from hyperspectral data with the high spectral resolution is presented. The intelligent selection of sensitive bands is a convenient way to build multispectral sensors for a specific application (in this article, the diagnosis of psylla leaves). Knowledge of disease-sensitive wavelengths can also help researchers analyze multispectral and hyperspectral aerial images captured by satellites or drones.Materials and MethodsA total number of 160 healthy and diseased leaves were scanned in 64 spectral bands between 400-1100 nm with 10 nm spectral resolution. A random forest algorithm was used to identify the importance of features in classifying the dataset into diseased and healthy leaves. After computing the importance of the features, a clustering algorithm was developed to cluster the most important features into six clusters such that the center of clusters was 50 nm apart. To transfer the hyperspectral dataset into a multispectral dataset, the reflectance was averaged in spectral bands within ±15 nm of each cluster center and achieved six broad multispectral bands. Afterwards a support vector machine algorithm was utilized to classify the diseased and healthy leaves using both hyperspectral and multispectral datasets.Results and DiscussionThe center of clusters were 468 nm, 598 nm, 710 nm, 791 nm, 858 nm, and 1023 nm, which were calculated by taking the average of all the members assigned to the individual clusters. These are the most informative spectral bands to distinguish the pistachio leaves infected by Psylla from the healthy leaves. The F1-score was 90.91 when the hyperspectral dataset (all bands) was used, while the F1-score was 88.69 for the multispectral dataset. The subtle difference between the F1-scores indicates that the proposed pipeline in this study was able to select appropriately the sensitive bands while retaining all relevant information.ConclusionThe importance of spectral bands in the visible and near-infrared region (between 400 and 1100 nm) was obtained to identify pistachio tree leaves infected with psylla disease. Based on the importance of spectral properties and using a clustering algorithm, six wavelengths were obtained as the best wavelengths for classifying healthy and diseased pistachio leaves. Then, by averaging the wavelengths at a distance of 15 nm from these six centers, the hyperspectral data (64 bands) became multispectral (6 bands). Since the correlation between the wavelengths in the near-infrared region was very high (more than 95%), out of the three selected wavelengths in the near-infrared region (710, 791, and 1023), only the 710-nm wavelength, which was closer to the visible region, was selected. The results of classification of infected and diseased leaves using hyperspectral and multispectral data showed that the degree of classification accuracy decreases by about 2% and if only 4 bands are used, the degree of accuracy decreases by about 3%.The results of this study revealed that the proposed framework could be used for selecting the most informative spectral bands and accordingly develop custom-designed multispectral sensors for disease detection in pistachio. In addition, we could reduce the dimensionality of the hyperspectral datasets and avoid the issues related to the curse of dimensionalitylity.
M. R. Zarezadeh; M. Aboonajmi; M. Ghasemi-Varnamkhasti; F. Azarikia
Abstract
IntroductionExtra Virgin Olive Oil (EVOO) is one of the most common and popular edible oils which is an important part of the Mediterranean diet. It is a rich source of sterol, phenol compounds and vitamins A and E. EVOO has useful effects on human body and significant reduction of cardiovascular diseases ...
Read More
IntroductionExtra Virgin Olive Oil (EVOO) is one of the most common and popular edible oils which is an important part of the Mediterranean diet. It is a rich source of sterol, phenol compounds and vitamins A and E. EVOO has useful effects on human body and significant reduction of cardiovascular diseases due to these benefits, EVOO is expensive so unfortunately adulteration in EVOO by mixing it with other cheap and low cost and low value oils such as canola, sunflower, palm and etc. is very common. Adulteration leads to health and financial losses and sometimes cause serious illness. Olive oil has various quality levels which depend on different factors such as olive cultivar, storage, oil extracting process etc.Materials and MethodsThere are numerous food quality evaluation and adulteration detection approaches which include destructive and non-destructive methods. Control sample (EVOO) was applied from "DANZEH food industry", Lowshan, Gilan Province. For ensuring that control sample is extra virgin, a sample was tested in "Rahpooyan e danesh koolak Lab." Tehran, Iran; according to "Institute of standards and industrial research of Iran" ISIRI number: 4091 and INSO 13126-2. Eight semi-conductor gas sensors "FIS, MQ3, MQ3, MQ4, MQ8, MQ135, MQ136, TGS136, TGS813 AND TGS822" applied in used olfaction machine. In this study there were 6 treatments: 1- Pure EVOO, 2- EVOO with 5% adulteration, 3- EVOO with 10% adulteration, 4- EVOO with 20% adulteration, 5- EVOO with 35% adulteration and 6- EVOO with 50% adulteration. Adulteration created with ordinary frying oil (including sunflower, canola, and maize oils). Each treatment prepared in seven samples and each sample test was repeated seven times. In this study, olfaction machine, a non-destructive, simple and user friendly System applied. As mentioned, the olfaction machine includes eight different sensors, so each test has eight graphs. Four features (1- Sensor output (mV) in start of odor pulse (refer to fig. 3) 2- Sensor output at the end of odor pulse 3- Average of sensor output during odor pulse and 4- Difference of sensor output at the end and start of start of odor pulse); So 32 features extracted and analyzed and finally effective sensors reported.Results and DiscussionHistogram and box plot of raw data showed that the data are not normal and need some preprocessing operations. Preprocessing facilitates data analyzing and classifying extracted features. After preprocessing, the standard data, divided into two classes: train data (70%) and test data (30%). Data classified with 4 different classifier models which include: K-nearest neighbors, support vector machine, artificial neural network and Ada-boost. Results showed that KNN method, with 89.89% and SVM with 86.52% classified with higher accuracy. Similarly, the confusion matrix showed the reasonable results of classifying operation. Also, three effective sensors in classifying determined TGS2620, MQ5 and MQ4 respectively, and on the other side, sensors such as MQ3 and MQ8 have the minimum effect on classifying so it is possible to remove these sensors from the sensor array without effective impress on results. This may cause decrease in the olfaction machine price and reduce analyzing time.ConclusionDue to increasing adulteration in foods, especially in olive oil and its significant effects on people's health and financial losses, a simple, cheap and non-destructive quality evaluation extended. Results showed that the olfaction machine with metal oxide semiconductor (especially including TGS 2620, MQ5 and MQ4 sensors) can use for classification and adulteration detection of extra virgin olive oil. Evaluation of this system's output leads to higher classification accuracy by using KNN and SVM method for olive oil classification and also fraud detection (5% adulteration).
N. Bagheri; H. Mohamadi-Monavar
Abstract
Fire Blight (FB) is the most destructive bacterial disease of pome fruit trees around the world. In recent years, spectrometry has been shown to be an accurate and real-time sensing technology for plant disease detection. So, the main objective of this research is early detecting FB of pear trees by ...
Read More
Fire Blight (FB) is the most destructive bacterial disease of pome fruit trees around the world. In recent years, spectrometry has been shown to be an accurate and real-time sensing technology for plant disease detection. So, the main objective of this research is early detecting FB of pear trees by using Visible-Near-infrared spectrometry. To get this goal, the reflectance spectra of healthy leaves (ND), non-symptomatic (NS), and symptomatic diseased leaves (SY) were captured in the visible–NIR spectral regions. In order to keep the important information of spectra and reduce the dimension of data, three linear and non-linear manifold-based learning techniques were applied such as, Principal Component Analysis (PCA), Sammon mapping and Multilayer auto-encoder (MAE). The output of manifold-based learning techniques was used as an input of the SIMCA (Soft independent modeling by class analogy) classification model to discriminate NS and ND leaves. Based on the results, the best classification accuracy obtained by using PCA on the 1st derivative spectra, with accuracy of 95.8%, 89.3%, and 91.6% for ND, NS, and SY samples, respectively. These results support the capability of manifold-based learning techniques for early detection of FB via spectrometry method.
S. Sabzi; Y. Abbaspour Gilandeh; H. Javadikia
Abstract
Introduction With increase in world population, one of the approaches to provide food is using site-specific management system or so-called precision farming. In this management system, management of crop production inputs such as fertilizers, lime, herbicides, seed, etc. is done based on farm location ...
Read More
Introduction With increase in world population, one of the approaches to provide food is using site-specific management system or so-called precision farming. In this management system, management of crop production inputs such as fertilizers, lime, herbicides, seed, etc. is done based on farm location features, with the aim of reducing waste, increasing revenues and maintaining environmental quality. Precision farming involves various aspects and is applicable on farm fields at all stages of tillage, planting, and harvesting. Today, in line with precision farming purposes, and to control weeds, pests, and diseases, all the efforts of specialists in precision farming is to reduce the amount of chemical substances in products. Although herbicides improve the quality and quantity of agricultural production, the possibility of applying inappropriately and unreasonably is very high. If the dose is too low, weed control is not performed correctly. Otherwise, If the dosage is too high, herbicides can be toxic for crops, can be transferred to soil and stay in it for a long time, and can penetrate to groundwater. By applying herbicides to variable rate, the potential for significant cost savings and reduced environmental damage to the products and environment will be possible. It is evident that in large-scale modern agriculture, individual management of each plant without using some advanced technologies is not possible. using machine vision systems is one of precision farming techniques to identify weeds. This study aimed to detect three plant such as Centaurea depressa M.B, Malvaneglecta and Potato plant using machine vision system. Materials and Methods In order to train algorithm of designed machine vision system, a platform that moved with the speed of 10.34 was used for shooting of Marfona potato fields. This platform was consisted of a chassis, camera (DFK23GM021,CMOS, 120 f/s, Made in Germany), and a processor system equipped with Matlab 2015 version. The video camera was installed in 60-centimeter height above the ground level. Therefore, all plants in the camera field of view (whether on the crops row or between the rows) were analyzed. This study conducted on 4 hectares of potato fields in Kermanshah–Iran (longitude: 7.03 E; latitude: 4.22 N). The most suitable color space for segmentation plants was HSV color space and most suitable channel of applying threshold was the H channel. In this study, features in two areas of color features, texture features based on gray co-occurrence matrix were extracted. Ultimately, 126 color features and 80 texture features were extracted from each object. In final six features among 206 features were selected. Results and Discussion Among 206 extracted features, six effective features including the additional second component of the YCbCr color space, green index minus blue in RGB color space, sum entropy in the neighborhood of 45 degree, diagonal moment in the neighborhood of 0 degree, entropy in the neighborhood of 45 degree, additional third component index in CMY color space were selected using hybrid ANN-PSO. This means that, two set features have the same effect over plants. The result shows that hybrid ANN-SAGA classified Centaurea depressa M.B, Malvaneglecta and Potato plant with 99.61% accuracy. This accuracy is high and this meant that 1. These plants have different 6 selected features, 2. The classifier is very powerful to classify. Conclusion 1. Plants with similar features make the classification process complicated and less accurate. 2. The presence of shadow on the plants’ leaves reduces the accuracy of the classification.
B. Jamshidi; S. Minaei; E. Mohajerani; H. Ghassemian
Abstract
In recent years, application of near-infrared spectroscopy (NIR) as a non-destructive technique combined with chemometric methods has been widely noticed for quality assessment of food and agricultural products. In chemometric methods, quality analyses are important issues which could be related to pattern ...
Read More
In recent years, application of near-infrared spectroscopy (NIR) as a non-destructive technique combined with chemometric methods has been widely noticed for quality assessment of food and agricultural products. In chemometric methods, quality analyses are important issues which could be related to pattern recognition. In this research, the feasibility of pattern recognition methods combined with reflectance NIR spectroscopy for non-destructive discrimination of oranges based on their tastes was investigated. To this end, both unsupervised and supervised pattern recognition techniques, hierarchical cluster analysis (HCA) and soft independent modeling of class analogies (SIMCA) were used for assessing the feasibility of variety discrimination and classification (according to their taste), respectively, based on the spectral information of 930-1650nm range. Qualitative analyses indicated that NIR spectra of orange varieties were correctly clustered using unsupervised pattern recognition of HCA. It was also concluded that supervised pattern recognition of SIMCA for NIR spectra of oranges provided excellent results of variety classification based on BrimA index at 5% significance level (classification accuracy of 98.57%). Moreover, wavelengths of 1047.5nm, 1502nm, and 1475nm contributed more than other wavelengths in discriminating two classes. Samples having the same BrimA index were also correctly classified with the high classification accuracy (95.45%) at 5% significance level. The discrimination power of wavelengths of 1475nm, 1583nm, and 1436.75nm were more than those for other wavelengths to achieve this classification. Therefore, reflectance NIR spectroscopy combined with pattern recognition methods can be utilized for determination of other attributes related to taste.
A. Rohani; H. Makarian
Abstract
With the rise of new powerful statistical techniques and neural networks models, the development of predictive species distribution models has rapidly increased in ecology. In this research, a learning vector quantization (LVQ) and multi layer perceptron (MLP) neural network models have been employed ...
Read More
With the rise of new powerful statistical techniques and neural networks models, the development of predictive species distribution models has rapidly increased in ecology. In this research, a learning vector quantization (LVQ) and multi layer perceptron (MLP) neural network models have been employed to predict, classify and map the spatial distribution of A. repens L. density. This method was evaluated based on data of weed density counted at 550 points of a fallow field located in Faculty of Agriculture, Shahrood University of Technology, Semnan, Iran, in 2010. Some statistical tests, such as comparisions of the means, variance, statistical distribution as well as coefficient of determination in linear regression were used between the observed point sample data and the estimated weed seedling density surfaces by two neural networks to evaluate the performance of the pattern recognition method. Results showed that in the training and test phases non significant different was observed between average, variance, statistical distribution in the observed and the estimated weed density by using LVQ neural network. While this comparisions was significant except statistical distribution by using MLP neural network. In addition, results indicated that trained LVQ neural network has a high capability in predicting weed density with recognition erorr less than 0.64 percent at unsampled points. While, MLP neural network recognition erorr was less than 14.6 percent at unsampled points. The maps showed that, patchy weed distribution offers large potential for using site-specific weed control on this field.