Plant leaf disease recognition using image-based machine learning
| Autor: | Radočaj, Petra |
|---|---|
| Přispěvatelé: | Krpić, Zdravko, Zorić, Bruno |
| Jazyk: | chorvatština |
| Rok vydání: | 2020 |
| Předmět: |
deep machine learning
TEHNIČKE ZNANOSTI. Računarstvo. Umjetna inteligencija classification dijagnosticiranje bolesti biljke konvolucijske neuronske mreže convolutional neural networks diagnosing plant diseases transfer learning duboko strojno učenje prijenosno učenje klasifikacija TECHNICAL SCIENCES. Computing. Artificial Intelligence |
| Popis: | U teorijskom dijelu ovoga rada opisan je problem raspoznavanja bolesti lista biljke uz prikaz mogućnosti primjene postupaka dubokog strojnog učenja. Razvoj modela za klasifikaciju bolesti potaknuo je unaprjeđenje automatiziranih sustava za otkrivanje biljnih bolesti pomoću vidljivih simptoma na lišću dok je sam proces izdvajanja značajki ugrađen u algoritam. Znanje o ovom procesu značajno bi doprinijelo području znanosti pri razumijevanju dijagnostičkih postupaka i detekciji bolesti. U eksperimentalnom dijelu rada proveden je postupak klasifikacije bolesti lista biljke na prerađenom GooLeNet i VGGNet modelu te MobileNet, InceptionV3 i InceptionResNetV2 mrežnim arhitekturama. Sve konvolucijske neuronske mreže trenirane su i testirane na odabranom skupu podataka koji uključuje bolesne i zdrave jedinke kukuruza podijeljene u četiri klase. Najpreciznije rezultate omogućio je InceptionResNetV2 model s 98.9% točnosti na odabranom skupu podataka. The problem of recognizing plant leaf diseases with a review of the current possibilities by applying deep machine learning is described in the theoretical part of this thesis. The development of disease classification models has prompted improvements of automated plant disease detection systems using visible symptoms, while the feature extraction is built in the algorithms. Knowledge of this process has contributed to the field of science in understanding diagnostic procedures and disease detection. In the experimental part of the thesis, the procedure of leaf diseases classification with the modified GoogleNet and VGGNet model MobileNet, alongside InceptionV3 and InceptionResNetV2 network architectures was performed. All convolutional neural networks were trained and tested on selected set data that included diseased and healthy maize images divided into four classes. The most accurate results were achieved by the InceptionResNetV2 model with 98.9% accuracy on the selected data set. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|