Figure 1:
Figure 2:
Figure 3:
Figure 4:
Figure 5:
Figure 6:
Figure 7:
Figure 8:
Figure 9:
Sugar content
| Name of the Apple | Apples | Sugar content (% Brix) |
|---|---|---|
| Kinnaur Apple | Apple 1 to Apple 4 | 10, 10, 15, 12.5 |
| Red Delicious NZ | Apple 1 to Apple 4 | 12.8, 13.8, 12.1, 10 |
| Red Delicious USA | Apple 1 to Apple 4 | 10, 12, 10, 10 |
| Royal Gala | Apple 1 to Apple 4 | 10, 12, 13, 13 |
| Washington Apple | Apple 1 to Apple 4 | 10, 12.5, 10, 10 |
Comparison of the proposed system with existing work
| System | Year | Technology | Application | Accuracy |
|---|---|---|---|---|
| Proposed system (AppleNet) | 2024 | CNN-based multispectral imaging | Grading Apples by Sweetness | 65% |
| Wang et al. [26] | 2023 | Near-infrared spectroscopy and multispectral imaging | Red Fuji Apples (defects, shape, sweetness) | 96.67% (defects), 94.67% (size & shape), near-perfect (sweetness) |
| Yan et al. [5] | 2023 | Enhanced YOLOv5s framework | Real-time Apple detection for picking robots | Recall: 91.48%, precision: 83.83%, F1: 87.49% |
| Iqbal et al. [22] | 2023 | Raman spectroscopy | Apple quality and safety | High sensitivity to quality and safety parameters |
| Liu et al. [10] | 2024 | Multi-band imaging and chemometric analysis | Rice seed quality and variety discrimination | Up to 94% for spectral-based distinctions |
Grading by Sweetness performance metrics of apple fruit
| Class name | Precision | Recall | F1-Score |
|---|---|---|---|
| Class-1-10% Brix sugar content | 0.6500 | 0.6500 | 0.6500 |
| Class-2-12% Brix sugar content | 0.6475 | 0.6475 | 0.6475 |
| Class-3-13% Brix sugar content 13 | 0.6500 | 0.6500 | 0.6500 |
| Class-4-15% Brix sugar content | 0.5638 | 0.5638 | 0.5638 |
| Accuracy | 65% | ||
| Misclassification rate | 0.3719 | ||
| Macro-F1 | 0.6278 | ||
| Weighted-F1 | 0.6281 |
Comparison of proposed technique with existing deep learning techniques
| Name of the network | No of layers | Type of network | Accuracy (%) | No of parameters | Error rate (%) | Features |
|---|---|---|---|---|---|---|
| AlexNet | 8 | CNN | 84.7 | 62 millions | 15.03 | Deeper |
| GoogleNet | 22 | CNN | 93.33 | 4 millions | 6.67 | Increased computational efficiency |
| DenseNet | 5 | CNN | 93.34 | 8 millions | 6.66 | Strong gradient flow |
| VGG-16 | 16 | CNN | 92.3 | 138 millions | 7.30 | Fixed-size kernels |
| VGG-19 | 19 | CNN | 92.3 | 143 millions | 7.30 | Fixed-size kernels |
| InceptionNet-v3 | 48 | CNN | 93.3 | 6.4 millions | 6.70 | Wider parallel kernels |
| AppleNet (Proposed method) | 7 | CNN | 65 (For sweetness) | 196 millions | 35 (For sweetness) | Low computation requirement |