Advancing mango leaf variant identification with a robust multi-layer perceptron model

Abstract

Mango, often regarded as the "king of fruits," holds a significant position in Bangladesh's agricultural landscape due to its popularity among the general population. However, identifying different types of mangoes, especially from mango leaves, poses a challenge for most people. While some studies have focused on mango type identification using fruit images, limited work has been done on classifying mango types based on leaf images. Early identification of mango types through leaf analysis is crucial for taking proactive steps in the cultivation process. This research introduces a novel multi-layer perceptron model called WaveVisionNet, designed to address this challenge using mango leaf datasets collected from five regions in Bangladesh. The MangoFolioBD dataset, comprising 16,646 annotated high-resolution images of mango leaves, has been curated and augmented to enhance robustness in real-world conditions. To validate the model, WaveVisionNet is evaluated on both the publicly available dataset and the MangoFolioBD dataset, achieving accuracy rates of 96.11% and 95.21%, respectively, outperforming state-of-the-art models such as Vision Transformer and transfer learning models. The model effectively combines the strengths of lightweight Convolutional Neural Networks and noise-resistant techniques, allowing for accurate analysis of mango leaf images while minimizing the impact of noise and environmental factors. The application of the WaveVisionNet model for automated mango leaf identification offers significant benefits to farmers, agricultural experts, agri-tech companies, government agencies, and consumers by enabling precise diagnosis of plant health, enhancing agricultural practices, and ultimately improving crop yields and quality.

Keywords: Agricultural AI; Mango leaf identification; MangoFolioBD dataset; Multi-layer perceptron (MLP); Noise-resistant image analysis; WaveVisionNet.

Fig.1

Medicinal leaf dataset sample images (collected online).

Fig. 2

Locations of mango orchards; (a) SOAS agro, (Feni) (b) Bangladesh Agricultural Research Institute (Rangpur), (c) Brac nursery (Demra), (d) Rajshahi division, (e) Munshiganj in Bangladesh selected for data collection.

Fig. 3

Geographic locations of mango orchards in Bangladesh selected for data collection. The map was adapted from the disaster and climate risk information platform (DRIP), developed by the Center for Environmental and Geographic Information Services (CEGIS) under the Government of the People’s Republic of Bangladesh. Reproduced with permission. ©Government of the People’s Republic of Bangladesh. All Rights Reserved http://drip.plancomm.gov.bd/BasicMap/MapViewer.

Fig. 4

Data collection process ensuring leaf sample accuracy and cleanliness. Visual examination, dusting, washing, and controlled drying techniques were employed.

Fig. 5

Sample Images of our MangoFolioBD dataset (primary data).

Fig. 6

Data augmentation and noise addition pipeline for our MangoFolioBD dataset.

Fig. 7

The overall technical path of this study and WaveVisionNet architecture.

Fig. 8

The working mechanism and receptive field of (a), ( b) dialated convolution (DC) and HDC with dilation rates of (3,3) and (2,2).

Fig. 9

Schematic overview of WaveVisionNet architecture for mango leaf categorization. The model enhances classification tasks by incorporating hierarchical feature extraction, attention mechanisms, and positional encoding, with self-attention applied initially and hybrid dilated convolution for enhanced receptive field.

Fig. 10

(a) ROC curves illustrating the performance of the top three models. (b) Confusion matrix depicting the assessment of our WaveVisionNet model across training iterations on the MangoFolioBD dataset.

Fig. 11

(a) Validation accuracy curves showcasing the performance of all models utilizing the WaveVisionNet model architecture on the public dataset. (b) Validation accuracy curves illustrating the performance of all models utilizing the WaveVisionNet model architecture on the MangoFolioBD dataset.

Fig. 12

Overall performance metrics evaluation on medicinal leaves dataset (a) (public) and MangoFolioBD dataset (b) (private).

Fig. 13

Robustness measurement of different metrics of our proposed WaveVisionNet model in with noise and without noise condition.

Fig. 14

(a–c) Output probabilities displayed by our WaveVisionNet model during the pretraining phase on the public dataset. (d–f) Output probabilities depicted by our WaveVisionNet model for accurate performance on the MangoFolio dataset, utilizing validation data directory.

Fig. 15

t-SNE visualizations of the MangoFolio dataset with 26 unique classes, highlighting insights into the model’s learned representation of high-dimensional data across different learning rates (0.5, 0.6, 0.7).

Fig. 16

Heatmaps generated using Grad-CAM technique from our proposed WaveVisionNet model. Heatmap corresponding to the first convolutional layer and final convolutional layer. These heatmaps highlight the regions in the input image that significantly influence the model’s predictions, providing insights into the model’s decision-making process.