Machine Learning (ML) is one of the disciplines of Artificial Intelligence (AI) with interesting applications in many fields, especially in agriculture. Our first interest was about the wine industry, a very important sector in Portugal (the country where I developed my thesis) and in Italy (my home country). We did a review of the state of the art to better understand the world of ML applied to this sector. One of the interesting applications we discovered was image recognition applied to disease detection by analyzing grape leaves. Then we looked at some applications of Convolutional Neural Network (CNN) and explored this topic in more detail. Applying this type of networks to a dataset composed of grapevine leaves with 3 different diseases and a subset with healthy grapevine leaves. We obtained very good results, always above 95% accuracy, using 4 different types of networks: ResNet50, GoogLeNet, InceptionV3, SqueezeNet, using the transefer learning technique. We wanted to highlight the limitations of the above techniques nowadays and support this thesis with some analytical results.
Il Machine Learning (ML) è una delle discipline dell'Intelligenza Artificiale (AI) con applicazioni interessanti in molti campi, specialmente in agricoltura. Il nostro primo interesse è stato riguardo l'industria del vino, un settore molto importante in Portogallo (il paese dove ho sviluppato la mia tesi) e in Italia (il mio paese d'origine). Abbiamo fatto una revisione dello stato dell'arte per capire meglio il mondo del ML applicato a questo settore. Una delle applicazioni interessanti scoperte è stato il riconoscimento delle immagini applicato al rilevamento delle malattie, analizzando le foglie dell'uva. Poi abbiamo analizzato alcune applicazioni di Convolutional Neural Network (CNN) e abbiamo approfondito più in dettaglio questo argomento. Applicando questo tipo di reti ad un dataset composto da foglie di vite con 3 diverse malattie e un subset con foglie di vite sane, abbiamo ottenuto ottimi risultati, sempre al di sopra del 95% di accuratezza, utilizzando 4 tipologie di reti differenti: ResNet50, GoogLeNet, InceptionV3, SqueezeNet, con la tecnica del transefer learning. Abbiamo voluto evidenziare i limiti delle suddette tecniche al giorno d'oggi e sostenere questa tesi con alcuni risultati analitici.
Machine learning in agriculture : convolutional neural network for disease detection in grape leaves
Balducci, Matteo
2020/2021
Abstract
Machine Learning (ML) is one of the disciplines of Artificial Intelligence (AI) with interesting applications in many fields, especially in agriculture. Our first interest was about the wine industry, a very important sector in Portugal (the country where I developed my thesis) and in Italy (my home country). We did a review of the state of the art to better understand the world of ML applied to this sector. One of the interesting applications we discovered was image recognition applied to disease detection by analyzing grape leaves. Then we looked at some applications of Convolutional Neural Network (CNN) and explored this topic in more detail. Applying this type of networks to a dataset composed of grapevine leaves with 3 different diseases and a subset with healthy grapevine leaves. We obtained very good results, always above 95% accuracy, using 4 different types of networks: ResNet50, GoogLeNet, InceptionV3, SqueezeNet, using the transefer learning technique. We wanted to highlight the limitations of the above techniques nowadays and support this thesis with some analytical results.File | Dimensione | Formato | |
---|---|---|---|
Tesi Matteo Balducci.pdf
non accessibile
Descrizione: Machine Learning in Agriculture: Convolutional Neural Network for Disease Detection in Grape Leaves
Dimensione
10.61 MB
Formato
Adobe PDF
|
10.61 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/179480