Deep learning-based approach for identification of diseases of maize crop

Abstract

In recent years, deep learning techniques have shown impressive performance in the field of identification of diseases of crops using digital images. In this work, a deep learning approach for identification of in-field diseased images of maize crop has been proposed. The images were captured from experimental fields of ICAR-IIMR, Ludhiana, India, targeted to three important diseases viz. Maydis Leaf Blight, Turcicum Leaf Blight and Banded Leaf and Sheath Blight in a non-destructive manner with varied backgrounds using digital cameras and smartphones. In order to solve the problem of class imbalance, artificial images were generated by rotation enhancement and brightness enhancement methods. In this study, three different architectures based on the framework of 'Inception-v3' network were trained with the collected diseased images of maize using baseline training approach. The best-performed model achieved an overall classification accuracy of 95.99% with average recall of 95.96% on the separate test dataset. Furthermore, we compared the performance of the best-performing model with some pre-trained state-of-the-art models and presented the comparative results in this manuscript. The results reported that best-performing model performed quite better than the pre-trained models. This demonstrates the applicability of baseline training approach of the proposed model for better feature extraction and learning. Overall performance analysis suggested that the best-performed model is efficient in recognizing diseases of maize from in-field images even with varied backgrounds.

Figure 1

Sample images of dataset (A) Healthy, (B) Maydis Leaf Blight, (C) Turcicum Leaf Blight and (D) Banded Leaf and Sheath Blight of Maize Crop.

Figure 2

Architecture of the proposed models: (A) Inception-v3_flatten-fc, (B) Inception-v3_GAP and (C) Inception-v3_GAP-fc.

Figure 3

Overall testing accuracies of the proposed models on different training-validation data configurations.

Figure 4

Confusion matrices of the proposed models on the 70–15 data configuration (A) Inception-v3_flatten-fc (B) Inception-v3_GAP and (C) Inception-v3_GAP-fc.

Figure 5

Average precision, recall and f1-score of the proposed models on 70–15 data configuration.

Figure 6

Computational behavior of the proposed models on 70–15 data configuration (A) Number of trainable parameters and (B) Training time per epoch.

Figure 7

Comparative analysis of classification performance of Inception-v3_GAP model with pre-trained models on 70–15 data configurations: (A) Classification Accuracy (B) Average Precision (C) Average Recall and (D) Average f1-score.

Figure 8

Comparison of computational behavior of Inception-v3_GAP model with pre-trained models on 70–15 data configuration: (A) Number of trainable parameters and (B) training time per epoch.

Figure 9

Brightness enhancement of sample images using four gamma (γ) values.

Figure 10

Effect of batch sizes in the model performance (A) batch size vs training time per epoch and (B) batch size vs testing accuracy of the model.

Figure 11

Effect of epochs in the testing accuracies.