Advancements in deep learning for accurate classification of grape leaves and diagnosis of grape diseases

Abstract

Plant diseases cause significant agricultural losses, demanding accurate detection methods. Traditional approaches relying on expert knowledge may be biased, but advancements in computing, particularly deep learning, offer non-experts effective tools. This study focuses on fine-tuning cutting-edge pre-trained CNN and vision transformer models to classify grape leaves and diagnose grape leaf diseases through digital images. Our research examined a PlantVillage dataset, which comprises 4062 leaf images distributed across four categories. Additionally, we utilized the Grapevine dataset, consisting of 500 leaf images. This dataset is organized into five distinct groups, with each group containing 100 images corresponding to one of the five grape types. The PlantVillage dataset focuses on four classes related to grape diseases, namely Black Rot, Leaf Blight, Healthy, and Esca leaves. On the other hand, the Grapevine dataset includes five classes for leaf recognition, specifically Ak, Alaidris, Buzgulu, Dimnit, and Nazli. In experiments with 14 CNN and 17 vision transformer models, deep learning demonstrated high accuracy in distinguishing grape diseases and recognizing leaves. Notably, four models achieved 100% accuracy on PlantVillage and Grapevine datasets, with Swinv2-Base standing out. This approach holds promise for enhancing crop productivity through early disease detection and providing insights into grape variety characterization in agriculture.

Deep Learning and Machine Learning for Automatic Grapevine Varieties Identification: A Brief Review

“…Abassi and Jalal [30], the only study published using an ML-based approach after 2023, achieved an accuracy of 83. 20%, although this result was lower than all the other studies that proposed DL-based approaches to classify the same dataset [7,[38][39][40][41][42]46]. In fact, the ability of DL-based approaches to automatically learn to extract useful features has been making exponential advances in computer vision since 2012, with the publication of Krizhevsky et al [23].…”

“…Detailed information on each architecture used can be found in Alzubaidi et al [87]. In a different way, Carneiro et al [47] and Kunduracioglu and Pacal [38] used Vision Transformers throught ViT [88], Swin-Transformers [89], MobileViT [90], Deit [91], and MaxViT [92] but followed the same learning strategy.…”

“…As in other fields of research, in grape variety identification, DL-based approaches outperformed ML-based approaches in both the number of publications and performance (see Figure 4). Among the included studies that used DL and images, the worst performance was achieved by Pereira et al [58] with an accuracy of 77.30%, while other studies achieved 100% [37][38][39][40]57]. In this respect, the architecture used by Pereira et al [58] was AlexNet, the first deep CNN proposed in the literature in 2012 [23].…”

“…Focusing on the studies by Pereira et al [58] and Nasiri et al [53], they used the pre-trained AlexNet and VGG-16 models, respectively. Despite good results in the past, ImageNet classification results show that there is a high probability of improving results if more recent pre-trained CNN architectures are used, as in Carneiro et al [45], Magalhães et al [43], or Kunduracioglu and Pacal [38]. In addition, there are newer CNN architectures that remain untested, for example MobileNetV3 [118], larger members of the EfficientNet family [84,119], ConvNext [120,121].…”

“…Carneiro et al [49] and Kunduracioglu and Pacal [38] explored transformers in the context of identifying grapevine varieties. Dosovitskiy et al [88] introduced the concept of self-attention in image classification with ViT models, with the aim of repeating the good results obtained by Transformer Networks in natural language processing.…”