AgroVision Machine Learning Repository

Dataset Description

The data that we used is image datasets that consist of 18 different objects, where 9 objects related to plant crop disease and the other 9 about fruit and vegetable ripeness [1],[2],[3],[4]. Each of the objects has different amount of class.

Here is the table overview about distribution of the data that our team used for this project.

Tabel 1. Dataset Distribution

No	Object Name	Category	Number of Class	Number of Training Set	Number of Test Set
1.	Apple (Apel)	Plant Crop Disease	4	8014	1943
2.	Bell Pepper (Paprika)	Plant Crop Disease	2	4033	962
3.	Cherry (Ceri)	Plant Crop Disease	2	4205	1574
4.	Corn (Jagung)	Plant Crop Disease	4	7665	1855
5.	Grape (Anggur)	Plant Crop Disease	4	7335	1825
6.	Peach (Persik)	Plant Crop Disease	2	3566	891
7.	Potato (Kentang)	Plant Crop Disease	3	5907	1442
8.	Strawberry (Stroberi)	Plant Crop Disease	2	3598	900
9.	Tomato (Tomat)	Plant Crop Disease	9	17195	4197
10.	Bell Pepper (Paprika)	Vegetable Ripeness	5	424	163
11.	Chile Pepper (Cabai)	Vegetable Ripeness	5	452	166
12.	Tomato (Tomat)	Vegetable Ripeness	4	1021	230
13.	Apple (Apel)	Fruit Ripeness	2	1814	237
14.	Banana (Pisang)	Fruit Ripeness	4	11793	1685
15.	Guava (Jambu)	Fruit Ripeness	2	846	145
16.	Lime (Jeruk Nipis)	Fruit Ripeness	2	1016	174
17.	Orange (Jeruk Nipis)	Fruit Ripeness	2	1672	239
18.	Pomegranate (Delima)	Fruit Ripeness	2	864	159

Modeling & Evaluation

For the modeling part, we mainly used four different kind of model architectures. The first one is the self-created architecture where we defined the detail of each models, such as the layers, neurons, etc. The other three is using the transfer learning approach that referred to the official Keras API documentation [5]. We used Xception, MobileNetV2, and DenseNet121. The comparison between these three models is as follows.

Tabel 2. Model Comparison of Xception, MobileNetV2, and DenseNet121

No	Model Name	Size (MB)	Parameters (M)	Depth
1.	Xception	88	22.9	81
2.	MobileNetV2	14	3.5	105
3.	DenseNet121	33	8.1	242

Here are the detailed metrics that we got after doing model development on each objects using several different approaches mentioned before.

Plant Crop Disease

Apple

Tabel 3. Metrics of Apple Crop Disease Object

No	Model	Accuracy	Loss	F1-Score
1.	MobileNetV2 (Non-Augmented 2)	1	0.000879	1
2.	Xception (Non-Augmented 1)	1	0.000626	1
3.	Xception (Non-Augmented 2)	1	0.000338	1
4.	DenseNet121 (Augmented 1)	1	0.000651	1
5.	DenseNet121 (Augmented 2)	1	0.000675	1
6.	MobileNetV2 (Augmented 1)	1	0.001026	1
7.	MobileNetV2 (Augmented 2)	1	0.000297	1
8.	Xception (Augmented 1)	1	0.000412	1
9.	Xception (Augmented 2)	1	0.002333	1
10.	DenseNet121 (Non-Augmented 1)	0.9994	0.002031	0.9995
11.	DenseNet121 (Non-Augmented 2)	0.9994	0.002697	0.9995
12.	MobileNetV2 (Non-Augmented 1)	0.9994	0.005050	0.9995
13.	Self-Created (Augmented 1)	0.9984	0.006339	0.9985
14.	Self-Created (Augmented 2)	0.9938	0.022232	0.9940
15.	Self-Created (Non-Augmented 2)	0.9788	0.135590	0.9792
16.	Self-Created (Non-Augmented 1)	0.9696	0.125299	0.9702

Bell Pepper

Tabel 4. Metrics of Bell Pepper Crop Disease Object

No	Model	Accuracy	Loss	F1-Score
1.	MobileNetV2 (Non-Augmented 1)	1	0.000050	1
2.	MobileNetV2 (Non-Augmented 2)	1	0.001048	1
3.	DenseNet121 (Augmented 2)	1	0.002050	1
4.	MobileNetV2 (Augmented 1)	1	0.000179	1
5.	MobileNetV2 (Augmented 2)	1	0.000706	1
6.	Xception (Augmented 1)	1	0.001602	1
7.	Xception (Augmented 2)	1	0.012214	1
8.	Self-Created (Augmented 1)	1	0.002179	1
9.	Self-Created (Augmented 2)	1	0.003563	1
10.	DenseNet121 (Non-Augmented 2)	0.9989	0.003093	0.998959
11.	Xception (Non-Augmented 1)	0.9989	0.002842	0.998959
12.	Xception (Non-Augmented 2)	0.9989	0.002909	0.998959
13.	DenseNet121 (Augmented 1)	0.9989	0.005044	0.998959
14.	DenseNet121 (Non-Augmented 1)	0.9979	0.008506	0.997918
15.	Self-Created (Non-Augmented 2)	0.9875	0.043474	0.987510
16.	Self-Created (Non-Augmented 1)	0.9823	0.079027	0.982315

Cherry

Tabel 5. Metrics of Cherry Crop Disease Object

No	Model	Accuracy	Loss	F1-Score
1.	DenseNet121 (Non-Augmented 1)	1	0.000451	1
2.	DenseNet121 (Non-Augmented 2)	1	0.000026	1
3.	MobileNetV2 (Non-Augmented 1)	1	0.000031	1
4.	MobileNetV2 (Non-Augmented 2)	1	0.000277	1
5.	Self-Created (Non-Augmented 2)	1	0.004201	1
6.	Self-Created (Non-Augmented 1)	1	0.000693	1
7.	Xception (Non Augmented 1)	1	0.000196	1
8.	Xception (Non Augmented 2)	1	0.0000009	1
9.	DenseNet121 (Augmented 1)	1	0.000431	1
10.	DenseNet121 (Augmented 2)	1	0.001912	1
11.	MobileNetV2 (Augmented 1)	1	0.000066	1
12.	MobileNetV2 (Augmented 2)	1	0.000272	1
13.	Xception (Augmented 1)	1	0.000049	1
14.	Xception (Augmented 2)	1	0.000067	1
15.	Self-Created (Augmented 1)	1	0.000573	1
16.	Self-Created (Augmented 2)	0.9993	0.002806	0.9993

Corn

Tabel 6. Metrics of Corn Crop Disease Object

No	Model	Accuracy	Loss	F1-Score
1.	Xception (Augmented 2)	0.9919	0.0412	0.9917
2.	DenseNet121 (Augmented 1)	0.9886	0.0397	0.9885
3.	DenseNet121 (Non-Augmented 2)	0.9876	0.0552	0.9871
4.	DenseNet121 (Augmented 2)	0.9870	0.0433	0.9867
5.	Xception (Augmented 1)	0.9870	0.0386	0.9866
6.	DenseNet121 (Non-Augmented 1)	0.9859	0.0488	0.9855
7.	MobileNetV2 (Augmented 2)	0.9859	0.0404	0.9856
8.	MobileNetV2 (Augmented 1)	0.9854	0.0573	0.9851
9.	MobileNetV2 (Non-Augmented 1)	0.9849	0.0717	0.9844
10.	Xception (Non-Augmented 2)	0.9849	0.0644	0.9844
11.	Xception (Non-Augmented 1)	0.9843	0.1160	0.9838
12.	MobileNet V2 (Non-Augmented 2)	0.9800	0.0926	0.9794
13.	Self-Created (Augmented 1)	0.9800	0.0571	0.9796
14.	Self-Created (Augmented 2)	0.9741	0.0857	0.9734
15.	Self-Created (Non-Augmented 2)	0.9498	0.3311	0.9480
16.	Self-Created (Non-Augmented 1)	0.9401	0.1670	0.9380

Grape

Tabel 7. Metrics of Grape Crop Disease Object

No	Model	Accuracy	Loss	F1-Score
1.	DenseNet 121 (Non-Augmented 2)	1	0.0412	1
2.	Xception (Non-Augmented 1)	1	0.0397	1
3.	DenseNet121 (Augmented 2)	1	0.0552	1
4.	MobileNetV2 (Augmented 1)	1	0.0433	1
5.	Xception (Augmented 2)	1	0.0386	1
6.	MobileNetV2 (Non-Augmented 1)	0.9994	0.0488	0.9994
7.	MobileNetV2 (Augmented 2)	0.9994	0.0404	0.9994
8.	Xception (Augmented 1)	0.9994	0.0573	0.9994
9.	DenseNet121 (Non-Augmented 1)	0.9989	0.0717	0.9989
10.	MobileNetV2 (Non-Augmented 2)	0.9989	0.0644	0.9988
11.	DenseNet121 (Augmented 1)	0.9989	0.1160	0.9989
12.	Xception (Non Augmented 2)	0.9983	0.0926	0.9983
13.	Self-Created (Augmented 2)	0.9873	0.0571	0.9875
14.	Self-Created (Non-Augmented 1)	0.9857	0.0857	0.9860
15.	Self-Created (Augmented 1)	0.9830	0.3311	0.9833
16.	Self-Created (Non-Augmented 2)	0.9819	0.1670	0.9824

Peach

Tabel 8. Metrics of Peach Crop Disease Object

No	Model	Accuracy	Loss	F1-Score
1.	DenseNet121 (Non-Augmented 1)	1	0.00094	1
2.	DenseNet121 (Non-Augmented 2)	1	0.00120	1
3.	DenseNet121 (Augmented 2)	1	0.00099	1
4.	MobileNetV2 (Augmented 1)	1	0.00196	1
5.	MobileNetV2 (Augmented 2)	1	0.00036	1
6.	Xception (Augmented 1)	1	0.00229	1
7.	Xception (Augmented 2)	1	0.00116	1
8.	MobileNetV2 (Non-Augmented 1)	0.9994520	0.00197	0.9988
9.	MobileNetV2 (Non-Augmented 2)	0.9988780	0.00277	0.9988
10.	Xception (Non-Augmented 1)	0.9988780	0.00159	0.9988
11.	Xception (Non-Augmented 2)	0.9988780	0.00405	0.9988
12.	DenseNet121 (Augmented 1)	0.9988780	0.00417	0.9988
13.	Self-Created (Augmented 2)	0.9977550	0.02150	0.9977
14.	Self-Created (Augmented 1)	0.9943880	0.03340	0.9943
15.	Self-Created (Non-Augmented 2)	0.9831650	0.13861	0.9831
16.	Self-Created (Non-Augmented 1)	0.9809200	0.12781	0.9809

Potato

Tabel 9. Metrics of Potato Crop Disease Object

No	Model	Accuracy	Loss	F1-Score
1.	DenseNet121 (Non-Augmented 1)	0.9972	0.0095	0.9972
2.	DenseNet121 (Non-Augmented 2)	0.9972	0.0063	0.9972
3.	Xception (Non-Augmented 1)	0.9965	0.0143	0.9965
4.	Xception (Non-Augmented 2)	0.9965	0.0202	0.9965
5.	DenseNet121 (Augmented 1)	0.9965	0.0156	0.9965
6.	MobileNetV2 (Augmented 1)	0.9965	0.0152	0.9965
7.	MobileNetV2 (Augmented 2)	0.9965	0.0103	0.9965
8.	MobileNetV2 (Non-Augmented 1)	0.9951	0.0162	0.9951
9.	MobileNetV2 (Non-Augmented 2)	0.9951	0.0318	0.9951
10.	DenseNet121 (Augmented 2)	0.9951	0.0118	0.9951

Strawberry

Tabel 10. Metrics of Strawberry Crop Disease Object

No	Model	Accuracy	Loss	F1-Score
1.	DenseNet121 (Non-Augmented 1)	1	0.00005605	1
2.	DenseNet121 (Non-Augmented 2)	1	0.00000031	1
3.	MobileNetV2 (Non-Augmented 1)	1	0.00000696	1
4.	MobileNetV2 (Non-Augmented 2)	1	0.00000024	1
5.	Xception (Non-Augmented 1)	1	0.00000055	1
6.	Xception (Non-Augmented 2)	1	0.00047643	1
7.	DenseNet121 (Augmented 1)	1	0.00002332	1
8.	DenseNet121 (Augmented 2)	1	0.00003996	1
9.	MobileNetV2 (Augmented 1)	1	0.00112955	1
10.	MobileNetV2 (Augmented 2)	1	0.00059112	1

Tomato

Tabel 11. Metrics of Tomato Crop Disease Object

No	Model	Accuracy	Loss	F1-Score
1.	DenseNet121 (Non-Augmented 2)	0.9916	0.0332	0.9916
2.	MobileNetV2 (Augmented 1)	0.9911	0.0327	0.9911
3.	DenseNet121 (Non0Augmented 1)	0.9909	0.0326	0.9909
4.	MobileNetV2 (Augmented 2)	0.9909	0.0309	0.9909
5.	MobileNetV2 (Non-Augmented 2)	0.9904	0.0417	0.9904
6.	DenseNet121 (Augmented 1)	0.9904	0.0288	0.9904
7.	DenseNet121 (Augmented 2)	0.9899	0.0383	0.9899
8.	MobileNetV2 (Non-Augmented 1)	0.9888	0.0343	0.9887

Fruit Ripeness

Apple

Tabel 12. Metrics of Apple Fruit Ripeness Object

No	Model	Accuracy	Loss	F1-Score
1.	Xception	0.9915	0.0292	0.9915
2.	MobileNetV2	0.9957	0.0181	0.9957
3.	DenseNet121	0.9915	0.0274	0.9915

Banana

Tabel 13. Metrics of Banana Fruit Ripeness Object

No	Model	Accuracy	Loss	F1-Score
1.	Xception	0.9679	0.0881	0.9678
2.	DenseNet121	0.9727	0.0916	0.9722

Guava

Tabel 14. Metrics of Guava Fruit Ripeness Object

No	Model	Accuracy	Loss	F1-Score
1.	Xception	1	0.0098	1
2.	MobileNetV2	1	0.0038	1
3.	DenseNet121	1	0.0099	1

Lime

Tabel 15. Metrics of Lime Fruit Ripeness Object

No	Model	Accuracy	Loss	F1-Score
1.	Xception	0.9482	0.1056	0.9471
2.	MobileNetV2	0.9827	0.0484	0.9825
3.	DenseNet121	0.9942	0.0399	0.9941

Orange

Tabel 16. Metrics of Orange Fruit Ripeness Object

No	Model	Accuracy	Loss	F1-Score
1.	Xception	0.9707	0.1048	0.9706
2.	MobileNetV2	0.9958	0.0378	0.9958
3.	DenseNet121	0.9916	0.0445	0.9916

Pomegranate

Tabel 17. Metrics of Pomegranate Fruit Ripeness Object

No	Model	Accuracy	Loss	F1-Score
1.	Xception	0.9937	0.0222	0.9926
2.	MobileNetV2	1	0.0145	1
3.	DenseNet121	0.9937	0.0290	0.9925

Vegetable Ripeness

Bell Pepper

Tabel 18. Metrics of Bell Pepper Vegetable Ripeness Object

No	Model	Accuracy	Loss	F1-Score
1.	DenseNet121 (Version 2)	0.9938	0.0415	0.9918
2.	MobileNetV2 (Version 1)	0.9783	0.0580	0.9717
3.	MobileNetV2 (Version 2)	0.9721	0.0726	0.9636
4.	DenseNet121 (Version 1)	0.9690	0.0871	0.9592
5.	Xception (Version 2)	0.9628	0.1439	0.9514
6.	Xception (Version 1)	0.9473	0.1538	0.9321

Chile Pepper

Tabel 19. Metrics of Chile Pepper Vegetable Ripeness Object

No	Model	Accuracy	Loss	F1-Score
1.	DenseNet121 (Version 1)	0.9484	0.1850	0.9417
2.	DenseNet121 (Version 2)	0.9484	0.1618	0.9454
3.	MobileNetV2 (Version 2)	0.9351	0.2705	0.9251
4.	MobileNetV2 (Version 1)	0.9251	0.2585	0.9113
5.	Xception (Version 2)	0.9217	0.2612	0.9124
6.	Xception (Version 1)	0.9134	0.2661	0.9024

Tomato

Tabel 20. Metrics of Tomato Vegetable Ripeness Object

No	Model	Accuracy	Loss	F1-Score
1.	DenseNet121 (Version 2)	0.9774	0.1355	0.9780
2.	Xception (Version 2)	0.9774	0.1618	0.9775
3.	MobileNetV2 (Version 2)	0.9718	0.1880	0.9722
4.	Xception (Version 1)	0.9718	0.2087	0.9710
5.	DenseNet121 (Version 1)	0.9690	0.1631	0.9700
6.	MobileNetV2 (Version 1)	0.9690	0.1950	0.9706

Notes

The difference between model names that ended with the letter "1" (e.g. "... Non-Augmented 1", "... Augmented 1", and "... Version 1") and the letter "2" (e.g. "... Non-Augmented 2", "... Augmented 2", and "... Version 2") is related to the layer that was used before the model output layer. Model names that ended with the letter "1" use GlobalMaxPooling2D for the last model layer before the output layer, while model names that ended with the letter "2" use GlobalAveragePooling2D.

References

[1] Suryawanshi, Yogesh; PATIL, Kailas; Chumchu, Prawit (2022), “VegNet: Vegetable Dataset with quality (Unripe, Ripe, Old, Dried and Damaged)”, Mendeley Data, V1, doi: 10.17632/6nxnjbn9w6.

[2] PATIL, Kailas; MESHRAM, Vishal (2021), “FruitNet: Indian Fruits Dataset with quality (Good, Bad & Mixed quality)”, Mendeley Data, V2, doi: 10.17632/b6fftwbr2v.2

[3] Roboflow Universe Projects, "Banana Ripeness Classification Dataset," Roboflow Universe, Roboflow, Dec. 2022. [Online]. Available: https://universe.roboflow.com/roboflow-universe-projects/banana-ripeness-classification.

[4] D. Singh, N. Jain, P. Jain, P. Kayal, S. Kumawat, and N. Batra, "PlantDoc: A Dataset for Visual Plant Disease Detection," in Proceedings of the 7th ACM IKDD CoDS and 25th COMAD, Hyderabad, India, 2020, pp. 249-253, doi: 10.1145/3371158.3371196.

[5] K. Team, "Keras documentation: Keras Applications," Keras.io, 2023. [Online]. Available: https://keras.io/api/applications/.