Issuu

https://doi.org/10.22214/ijraset.2022.41468

April 2022

Leaf Disease Remedy Using CNN Algorithm Jahangir Jepee Mehta1, Furqaan Ahmad Wani2, Aamir Ashraf Ahangar3, Kanwaljeet Kaur4, Najmusher H5 1, 2, 3, 4

Student, 5Assistant Prof, Dept. of CSE, HKBKCE, Bangalore, India

Abstract: The proposed method aids in the diagnosis of plant diseases as well as the provision of medicines that may be employed as a defensive machine against them. The file collected from the web is correctly separated, and the various plant types are recognized and named again to produce a suitable record. A test file including several plant ailments is then obtained, which is used to assess the project's accuracy and confidence level. We'll next train our classifier with training data, and the result will be expected with maximum accuracy. We employ a Deep Convolutional Neuronic network (CNN), which consists of many layers for an estimate. A newly designed drone prototypical is also being developed that can be used to provide live updates of huge farming lands. The drone will be equipped with a highresolution photographic camera that will capture the image of the plants, which will be used as a contribution to the software, which will determine whether the plant is healthy or not. We reached a 78 percent accuracy level with our programming and training model. Our programmer provides us with the identity of the plant species, as well as the confidence level of the species and the medicine that may be used to treat it. Keywords: Machine Learning, Leaf Disease, Remedy, CNN I. INTRODUCTION The right assumption for stocks can incite huge benefits for the vendor and the trader. Each now and once more, it is drawn out that figure is tempestuous rather than discretionary, which would not kid about this can be expected by means of carefully taking apart the authentic background of independent monetary trade. AI is a capable way to deal with show these cycles. It predicts market regard close to the considerable worth, thusly extending the exactness. The Foundation of Machine learning in the district of the stock forecast has addressed various explores as a consequence of its capable and exact assessments. Information assortment is vital in this perspective as we depend on the information, particularly as we probably are aware of anticipating the financial exchange information. Along these lines, information should be captured based on the model. For example, if we are anticipating intraday trading, we must capture information consistently intensities (RDI). The algorithm is applicable not just to greenhouse-grown crops, but also to agriculturally grown crops. II. LITERATURE SURVEY Earlier articles described several systems for detecting pests such as aphids, whiteflies, thrips, and others, as well as diverse implementation methods, as demonstrated and discussed below. A conscious vision system was presented that included picture recognition, understanding, and knowledge-based techniques. Only the adult phase of the whitefly is detected, and the number of insects on a single leaflet is counted. As a test dataset, they utilized 180 photographs. They looked at 162 photographs in all, each of which had 0 to 5 snowfly bugs. For test photographs with no snowflies (class 1), at least one snowfly (class 2), and the whole test set, they calculate both the False-Negative Rate (FNR) and the False-Positive Rate (FPR). Expand its use of image analysis technologies and techniques to identify pests in a controlled environment such as a greenhouse. Three sorts of character qualities were reviewed and researched to distinguish the three different types of insect species, whiteflies, pests, and caterpillars: mass, morphological feature (a type of border), and color components. Encourage the use of video analysis to discover pests early in greenhouses. Their objective was to create a decision-making system using video camera data. They developed algorithms to detect only two types of bio-aggressors: whiteflies and aphids. By detecting whitefly eggs and studying whitefly activity, the device was sensitive enough to detect low infection stages. The planned pest detection system includes four points named reduction in noise, color conversion, counting of whiteflies, and segmentation. The comparative variation in pixels is automatically given a different algorithm name. Contextual parameter tweaking for adaptive picture segmentation has been included, allowing for more effective tuning of algorithm parameters in response to fluctuations in leaf shade and divergence. Presents an automated approach for classifying the primary mediators that harm soybean leaflets, such as beetles and caterpillars, using SVM classification, which is extensively used in feed-forward network programming. It also lacks essential uniqueness detection, necessitating training on known outcomes for feed-forward networks.

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 |

1148

International Journal for Research in Applied Science & Engineering Technology (IJRASET) ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538 Volume 10 Issue IV Apr 2022- Available at www.ijraset.com With a 96 percent accuracy rate, the system was tested on over 100 photos of whitefly pests. Proposed a novel pest recognition and positioning approach based on the binocular stereo to obtain insect location information, which was utilized to guide the robot to spray pesticides. III. METHODOLOGY A. Using a Support Vector Machine, early Intervention of pests on Plants is Possible This project is about a new form of pest system early detection. A digital camera is used to capture images of pest-affected leaves. The pest pictures on the leaves are processed to create a gray-colored image, which is then used to detect pests on the leaves using image segmentation and image classification algorithms. The image is sent to the analysis algorithm, which calculates the quality. This system uses combined image recognition and soft computing techniques. The pests are detected using image processing, and the detection is done over a large population using soft computing technologies. The photographs were captured with a digital camera with a 12-megapixel resolution and a 24-bit color depth. Following that, the photos are transmitted to a computer and represented using Open-CV software. For the separation of pests on leaves, the RGB picture is separated using a blob-like method. The sliced leaf portion is now being studied in order to estimate the insect population in the field. The pest kinds are categorized using the Support Vector Machine classifier. The Open-CV coding is converted into an HDL coder in the FPGA kit as well. The input picture is downloaded into memory in the FPGA. It retrieves a picture from storage, modifies it, and provides results on the monitor. In Open-CV, we created a software procedure. In this case, many neural network classifiers are used for training and testing. The following are some methods for classifying texture features. 1) K-nearest Neighbor: K-nearest neighbor is a term used to describe a person's closest A classifier calculates the shortest distance between a point and other points to identify which class it belongs to. The goal is to find the smallest distance between both the SQL queries sample and each develop a variety and pick the closest neighbor. 2) Radial Basis Function: An RBF is now a genuine role whose percentage is solely determined by the distance to the origin. Euclidean distance is the most often used measurement standard. RBFs are networks in which the range between the input sequence and a prototype sequence determines the initiation of hidden units. B. Artificial Neural Networks A form of Machine Learning method is Artificial Neural Networks (ANN) which have been increasingly popular in the current centuries. Multilayer Perception (MLP) is a simple type of ANN that uses backpropagation to change the weights during training [16]. Other neural network versions have recently gained popularity in texture categorization. PNN (Potentiality Neural Network): It is a parallel and distributed processor that is resultant from the (RBF) Radial Basis Function system and takes a natural inclination to store experience information. The activities are structured into a multilevel circuit configuration with 4 levels: source level, sequence level, processing level, and product-level in PNN, which is a modification of a computational algorithm-based gaussian discriminant analysis. C. Support Vector Machine SVM stands for "non-linear classifier" as well as being a pattern recognition technology that is still relatively new. SVM is used to tackle a variety of pattern classification, including texture classification. Only two classes were considered when SVM was developed. This is performed by increasing the margin of the hyperplane. Support vectors are the values nearest to the boundary that was chosen to determine the hyperplane. Multiclass classification may be used to solve an issue and is essentially made up of several two-class SVMs, either utilizing one-versus-all or one. Another aspect is the kernel function, which projects non-linearly distinguishable data from a low-dimensional area to a higher-dimensional space, allowing them to become separable in that space as well. The suggested method begins with capturing the data from the digital image and extracting the characteristics. The characteristics of the sample are saved in the database once it is photographed using a digital camera. Low-frequency background noise is removed by preprocessing photographs. The intensity of each picture particle is normalized. It improves the aesthetic appeal of photographs. Improve the ability to manipulate datasets. It's a method for improving data pictures before they're processed computationally. The caveat is that if not employed appropriately, enhancing methods might accentuate visual imperfections or possibly result in data loss. The first stage in preprocessing is to get an image as input, which must subsequently be upgraded. The Image database is then transformed into an image pixel in order to clearly identify pests on leaves. Filtering methods can be used to fulfill the noise removal purpose.

©IJRASET: All Rights are Reserved | SJ Impact Factor 7.538 | ISRA Journal Impact Factor 7.894 |

1149

International Journal for Research in Applied Science & Engineering Technology (IJRASET) ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538 Volume 10 Issue IV Apr 2022- Available at www.ijraset.com Filtering by average: The full image is scanned across the 3x3 sub-region. The average value replaces the center pixel at each location. The 3x3 sub-region is examined over the full picture as part of the median filtering process. The median value replaces the center pixel at each point. For both the average and standard filters, the PSNR score is computed. One of the filtered pictures is chosen for further processing based on this PSNR value. The PSNR score for mean filtering is 23.78, whereas the PSNR score for an average filter is 12.89. The finer the accuracy of the condensed or reconstructed image, the higher the PSNR. As a result, the mean screening is used in the next step. In general, image separation is described as the method of dividing a picture into homogeneous groups, with each region being homogeneous but no two neighboring regions being homogeneous. Picture segmentation is used to distinguish the many sections in an image that have specific relevance. These areas do not cross each other. Blob detection aids in the identification of regions of the image for further analysis. It is used to indicate the existence of multiples of a certain sort of object. Separate the interesting things (whiteflies) from the busy backdrop (leaves). Color, form, and texture are common image characteristics. There are three phases to feature extraction when using the General Based Voting method: HOG is for Histogram Oriented Gradient, GMM stands for Gaussian Mixture Model, and Gabor stands for Gabor Feature. HOG is a feature descriptor that may be used to detect objects. For texture analysis, the Gaussian mixture system is proposed. In the original image, the Gabor Feature calculates the association among groups of two pixels. The picture in this suggested work can be broken into tiny parts. The three stages are then included in each block. For detecting the distribution, HOG is utilized. D. Back Propagation Network The input vector, hidden vector, and output vector are the three sections of a conventional BP network. The collected weight value among nodes is used to link the three pieces. The BP network's most notable feature is that the system weight rate achieves potentials via the quantity of inaccuracy squares among the web outcome and the model outcome, and the system structure's weight value is then continually changed. In a picture, it is the color ratio. The form of bugs present in a picture is detected using GMM. Pests' orientation may be determined using the Gabor feature. The obtained features are then given to the classifiers as input. There are three different types of analyzers that have been used to determine which one produces the best results. Some bugs in a picture are missed by back propagation and feed-forward classifiers. However, SVM produces a superior outcome. SVM stands for "non-linear classifier" and is a relatively recent ML technique. Many patterns appreciation tasks, as well as texture sorting, are explained by means of SVM. Support Vector Machine was formed with just two modules in observance. This is skilled by maximizing the hyperplane's boundary. Support directions are the values nearest to the perimeter that was chosen to determine the hyperplane. Multiclass classification is effectively made up of numerous two-class SVMs, either using yet another or only one. Another aspect is the kernel function, which projects non-linearly recoverable data from a lowdimensional area to a higher-dimensional area, allowing them to become separable in that space as well. It is used to identify pests on leaves as well as provide information about pest types. It displays the results of a variety of pests. It then supplies a treatment to carry over for pest control. Finally, the parameters are loaded into a Support Vector Machine encoder, which allows us to discriminate between pests and leaves with precision. This is a critical step in identifying pests and applying the appropriate treatments. E. Organization of Fungal Disease Indicators affected by Cereals by means of Color Texture Features The detection and categorization of visual symptoms influenced by fungal illness using Artificial Neural Network and Support Vector Machine are described in this research. In this project, color photos of fungal infection symptoms caused by grains including maize, jowar, and wheat are used. The research takes into account a variety of symptoms caused by fungal disease, including leaf rot, bacterial leaf, dusty fungus, leaf rust, and smut. Preprocessing, segmenting, and extracting characteristics from disease-affected areas are all done with the created methods. The k-means segmentation approach is used to separate the impacted regions. Color texture characteristics are retrieved from the afflicted areas and fed into SVM and ANN processors as inputs. Color Co-occurrence Matrix is used to analyze texture. To categorize picture samples, tests are run. An ANN classifier may achieve classification accuracies of 68.5 percent to 87 percent. Using the SVM classifier, average classification accuracies climbed to 77.5 percent and 91.16 percent, respectively.

1150

International Journal for Research in Applied Science & Engineering Technology (IJRASET) ISSN: 2321-9653; IC Value: 45.98; SJ Impact Factor: 7.538 Volume 10 Issue IV Apr 2022- Available at www.ijraset.com IV. CONCLUSION The suggested approach was created with the interest of farmers and the agricultural industry in mind. The proposed method can identify illness in plants as well as give treatment options. Plant health may be improved by having a thorough understanding of the illness and its treatment. The suggested system is written in Python and has a 78 percent accuracy. The accuracy and speed of processing can be improved by using Google's GPU. Drones may be equipped with the system, allowing for aerial observation of agriculture fields. REFERENCES [1] [2]

Mrunalini R. et al., An application of K-means clustering and artificial intelligence in pattern recognition for crop diseases ,2011. S.Raj Kumar , S.Sowrirajan,‖ Automatic Leaf Disease Detection and Classification using Hybrid Features and Supervised Classifier‖, International Journal of Advanced Research in Electrical, Electronics and Instrumentation Engineering, vol. 5, Issue 6,2016. [3] Tatem, D. J. Rogers, and S. I. Hay, ―Global transport networks and infectious disease spread, ‖ Advances in Parasitology, vol. 62, pp. 293–343, 2006. View at Publisher · View at Google Scholar · View at Scopus. [4] J. R. Rohr, T. R. Raffel, J. M. Romansic, H. McCallum, and P. J. Hudson, ―Evaluating the links between climate, disease spread, and amphibian declines, ‖ Proceedings of the National Academy of Sciences of the United States of America, vol. 105, no. 45, pp. 17436–17441, 2008. View at Publisher · View at Google Scholar · View at Scopus. [5] T. Van der Zwet, ―Present worldwide distribution of fire blight, ‖ in Proceedings of the 9th International Workshop on Fire Blight, vol. 590, Napier, New Zealand, October 2001. [6] H. Cartwright, Ed., Artificial Neural Networks, Humana Press, 2015. Steinwart and A. Christmann, Support Vector Machines, Springer Science & Business Media, New York, NY, USA, 2008. View at MathSciNet. [7] S. Sankaran, A. Mishra, R. Ehsani, and C. Davis, ―A reviewof advanced techniques for detecting plant diseases, ‖ Computers and Electronics in Agriculture, vol. 72, no. 1, pp. 1–13, 2010. View at Publisher ·View at Google Scholar · View at Scopus. [8] P. R. Reddy, S. N. Divya, and R. Vijayalakshmi, ―Plant disease detection technique tool—a theoretical approach,‖ International Journal of Innovative Technology and Research, pp. 91–93, 2015. View at Google Scholar. [9] A.-K. Mahlein, T. Rumpf, P. Welke et al., ―Development of spectral indices for detecting and identifying plant diseases, ‖ Remote Sensing of Environment, vol. 128, pp. 21– 30, 2013. View at Publisher · View at Google Scholar · View at Scopus. [10] N. R. Deepak and S. Balaji, "Performance analysis of MIMO-based transmission techniques for image quality in 4G wireless network," 2015 IEEE International Conference on Computational Intelligence and Computing Research (ICCIC), 2015, pp. 1-5, doi: 10.1109/ICCIC.2015.7435774.

1151