machine learning a a tool for disease detection and diagnosis
- 1. Doctoral seminar-I Machine Learning : Modern Tool in Plant Disease Diagnosis Presented by - Prince Kumar Gupta Ph.D (Ag.) student G.B.P.U.A.& T, Pantnagar
- 2. Contents Importance of Plant Disease Diagnosis of Plant Pathology Artificial Intelligence Introduction of Machine Learning Steps in M.L algorithm Applications of M.L Deep Learning M.L based Mobile APPs Case studies Future aspects and Conclusion
- 3. Importance of Plant Disease Plant disease causes substainable looses in yield, leading to huge economic losses. Not only hamper the productivity, but also affect the human and livestock health It exploit plant, resulting in low productivity which acts as threat for food and nutritional security. Rekha et al.,2017
- 4. Diagnosis of plant disease is to identify the disease and to determine the causal agent (Pathogen). Diagnosis of Plant Disease Traditional method of identification –Visual examination.
- 5. Laboratory 1 Soil analysis and measurement of pH 2 Biochemical method 3 Microscopic examination 4 Serological methods 5 Use of methods rely on DNA Direct PCR FISH ELISA Immunoflurescence Flow cytometry Indirect Thermography Florescence imaging Hyperspectral Technique Artificial intelligence Current Prospectives: Fang et al.,2015
- 6. Artificial Intelligence 1956, John Mc Carty ‟The science and engineering of making intelligent machine” System able to perform task that require human intelligent- Visual perception, Speech recognition, face recognition and Translation of languages AI: Not a system , AI – Implemented in system.
- 9. Key Terminologies Data into information Problem solving tool Combination of CS engineering and Statistics Interpretet data and act on it Enhance performance using past experiences
- 10. what? Learning is any process by which a system improves performance from experience Term Machine learning was coined in 1959 by Arthur Samuel. Science of making computer learn and act like human by feeding data and information without being explicitly programmed The subset of Artificial Intelligence that work automatically or give the instructions to a particular system to perform a action and enhanced performance from past experience. Goal : Understand the structure of given data and fit data into models that can be understood and utilized by the people. Alex et al.,2008
- 11. Traine the system by feeding data Analyse the data ResultMachine learning makes prediction and decision based on past data System learn (algorithm) Data input (for testing) images
- 12. Why machine learning? Control: Varieties of fungicides are available to control diseases. Crop disease (presence of disease mainly reflected by symptoms on leaves )
- 13. Need: Automatic, accurate and less expensive Machine Vision System for detection of diseases from the image and to suggest a proper management For accurate plant disease diagnosis and effective chemicals- a difficult task - requires experts advise - time consuming and expensive
- 14. Why now ? Flood of available data Increasing computational power Progress in available algorithms & Theory developed by researchers Increasing support from industries Why now ?
- 15. Steps in Machine Learning 1. Collection of data (images) 2. Tabulation 3. Data Prepration High quality Training of data Clean dataOrganize data Elimination Create high quality data
- 16. Steps in Machine Learning 4. Data Input 5. Data Processing 6. Data Output First stage - raw data begins - form usable information. Machine learning algorithm In form of images, text etc.
- 17. 8. Data Storage 7. Prediction Alex et al.,2008 Database
- 18. Types of Machine Learning algorithm Supervised learning 1. Training model 2. Both input (image) +Visual symptoms 4. Naire Baye algorithm 3. Classsification Learning algorithm Model Blast of rice having specific disease symptoms Linear Regression Nearest Neighbor Guassian Naive Bayes Decision Trees Support Vector Machine (SVM) Random ForestResult
- 19. Unsupervised learning 1. Only input (images) 3. K-mean algorithm 2. Clustering (on the basis of features) result
- 20. Reinforcement learning algorithm Data (image) Model using algorithm Result save in database For detection test If result given by model is same as data already store in model database then its means model is accurant and rewarded as improving efficiency .
- 22. Application in Agriculture Data aggregation- water uses in irrigation. Signal sensor is used viz; GSM and soil sensor. Soil sensor- provides moisture, and water level regasrding informations. GSM- Collects weather data information
- 23. Through machine learning software, drone can adopt to learn to navigate better research and evaluates crop and livestock based on the data that already feed.
- 24. Robocrop: smart robert for picking fruits Crop monitoring system
- 25. Animal identification and Health monitoring.
- 26. Alibaba and Dekon By observing movement of pig/day - evaluate health of particular pig. By recording sound of pig cough they monitor the spread of disease. By use of machine learning, the mortality of piglet reduced.
- 27. Controlling Greenhouse Climate (Liakos et al.,2018)
- 28. Handling huge amount of data Complex Problems Feature Extraction
- 29. Deep learning.... Deep learning is a subset of machine learning Based on Artificial Neural Network Inspired by the Biological Neural Network Different algorithm are used most common Convolution Neural Network (CNN)
- 30. Convolutional Neural Network CNN commonly applied to analyzing visual imagery using neurons. Applicable in image recognition and image classification using filters. Convolution Pooling Scanning
- 31. Input (image number, width, height) identfy by machine in form of pixel (having image value)and these are then scan by filter (convolution layer). Filter are mainly of different type: color, edge,corner filter And of different size with some random values. convolution convolutionpooling pooling
- 40. Methodology Collection of image: Healthy: 95, Bscterial blight: 125, Blast: 170 , Sheath rot: 110 and Brown spot: 150.
- 41. Pre processing stage…….. Image are resize and cropped into pixel (300 x 450) Only leaf portion with the disease part is present in image.
- 42. K-mean clustering based segmentation Feature extraction Color Texture
- 44. Accuracy (%) when model implement…… Normal : 90.57; Bacterial blight: 95.78; Blast: 98.9; Brown spot: 94 and Sheath rot: 92
- 47. Layer Type Filter size Output size L 1 CONV 3X3 128X128X32 POOL 2X2 64X64X32 L 2 CONV 4X4 61X61X64 POOL 2X2 64X64X32 L 3 CONV 1X1 30x30x128 POOL 2X2 15x15x128 Training accuracy (80%) Test accuracy (20%) 99.21 % 99.32% Precision performance of the model Healthy 1.0 Septorial leaf blight 0.99 Frogeye 0.99 Downy mildew 0.98 Result
- 48. Linking machine learning with DSS (Decision Support System) for better advisory service for farmer. Image processing and spreading the usage of the model by training it for plant disease recognition on wider land areas. Severity of the disease Price list for the pesticides
- 49. Conclusion….. Automatic, Speed and Accuracy. Innovative, efficient & fast interpreting algorithms help in plant disease detection Bridge the gap between experimentation and real life application.