Fig 1.
shows the difficulties and barriers facing research presently.
Fig 2.
Overall description of work.
Fig 3.
Illustrate the overall standard steps of data collection and analysis procedures.
Fig 4.
Overall description of data acquisition sources, dataset types, preprocessing, methodology, and evaluation.
Table 1.
Apple’s datasets distributions.
Fig 5.
The Apple Fruit Varieties Collection (AFVC) distributions through 85 classes.
Fig 6.
The Apple Fruit Varieties Collection (AFVC) measurement was split through 85 classes; the overall training was 26,775, and the testing was 2,975 samples.
Fig 7.
The Apple Fruit Quality Categorization (AFQC).
Fig 8.
The Apple Diseases Extensive Collection distributions with fresh and rotten apples.
(ADEC) distributions with seven classes.
Fig 9.
The proposed optimized Apple orchard model vision transformers architecture, patches, vectorizations, Transformer Encoder Network, and MLP head.
Fig 10.
The MFCE loss function curves with.
Fig 11.
The curves of the MFCE loss function with varying values of dv.distinct γ values.
Table 2.
The environment configuration for the experiment.
Table 3.
The accuracy (%) of five deep-learning models on the AFVC, ADEC, and AFQC datasets.
Fig 12.
Accuracy, Loss, and validations of the pertained InceptionV3 on AFVC (85) dataset.
Fig 13.
Accuracy, Loss, and validations of the pertained InceptionV3 on ADEC (7) dataset.
Fig 14.
Accuracy, Loss, and validations of the pertained InceptionV3 on AFQC (2) datasets.
Table 4.
The assessment outcomes using the apple dataset sets for eight deep neural networks. Values are in percentages.
Fig 15.
The proposed method’s accuracy, average, weight, precision, recall, and f1-score on AFVC (85).
Fig 16.
Accuracy, Loss, and validations of the OAOM-VT on AFVC (85) dataset.
Fig 17.
The proposed method OAOM-VT on ADEC datasets: (A) accuracy and loss validations; (B) confusion matrix; (C) precision, recall, and f1-score.
Fig 18.
The proposed method OAOM-VT on AFQC datasets: (A) accuracy and loss validations; (B) confusion matrix; (C) precision, recall, and f1-score.
Table 5.
The result of the OAOM-VT model on various fruit categorization datasets with six, five, and four classes. Values are in percentages.
Fig 19.
The outperforms of the proposed method OAOM-VT with baselines on FAF (6C), AFD (5C), and F-AV (4C).