Recent Advancements in Fruit Detection and Classification Using Deep Learning Techniques

Table 1

Summary of the deep learning models applied to fruit detection.

Ref/year	DL model	Dataset	Dataset partition	Accuracy
[71] 2016	Faster R-CNN	TL + Field Farm	82% Train, 18% Test	0.83 F1-s
[41] 2017	Faster R-CNN	Orchard	2268 Train, 482 Test	>0.9 F1-s
[72] 2017	IN-ResNet	Personalized	24000 Train, 2400 Test	91% - 93
[41] 2017	VGG-16	Orchard	2268 Train, 482 Test	95%
[73] 2018	CNN	Kiwifruit	70% Train, 30% Test	89.29%
[74] 2019	YOLO V3	PT + WGISD	—	—
[75] 2019	DAN	Fruit 360	70% Train, 30% Test	91%
[76] 2019	Faster R-CNN + Iv2	Cherries	60% Train, 20% Val, 20% Test	85%
[77] 2019	E-Net	Fruit 360	80% Train, 20% Test	93.7%
[78] 2019	SS-CNN	Apple/Pears Orchard	—	+90%
[79] 2019	M-YOLO	PT + Mango Orchard	1300 Train, 130 Validation, 300 Test	0.97 F1-s
[80] 2019	M-Net	Mango Orchard	1300 Train, 130 Validation, 300 Test	73.6%
[81] 2019	M-RCNN + RetinaNet + FPN	Strawberry Dataset	2000 Train, 100 Test	95.78%
[82] 2019	Faster R-CNN + VGG-16	Kiwifruits	70% Train, 30%	Test -
[82] 2019	MMF MKV2	Kiwifruits	70% Train, 30% Test	±90%
[83] 2019	MVGG-16	Guava	80% Train, 20% Test	98.3%
[84] 2019	MVGG-16	Date Fruit	80% Train, 20% Test	98.59%
[83] 2019	MGNet	Guava	80% Train, 20% Test	94.8%
[84] 2019	AlexNet	Date Fruit	80% Train, 20% Test	99.01%, 97.01%
[85] 2019	ResNet	Strawberry	80% Train, 20% Test	94%
[86] 2019	MR-CNN + RNet-101	Orange	60% Train, 20% Validation, 20% Test	97.53%
[87] 2020	YOLO V3	PT + WGISD	Pretrained + 300 Train, 60	Test 97.3%
[88] 2020	YOLO V2	Mango + WGISD	300 Train, 60 Test	96.1%
[87] 2020	YOLO V2	Mango + WGISD	300 Train, 60 Test	95.6%
[89] 2020	YOLO V4	Banana Orchard	835 Train, 209 Validation, 120 Test	99.29%
[90] 2020	IM-R-CNN	Apple	368 Train, 120 Test	97.31%PR
[88] 2020	M-YOLOV3	Mango Orchard	1300 Train, 130 Validation, 300 Test	94% F1-s
[91] 2020	YOLO V4 + U-Net	Litchi Fruits	—	100%
[17] 2020	RetinaNet-FPN V4	Strawberry	80% Train, 20% Test	—

TL = transfer learning, F1-s: F1-score, PR: precision rate, PT: pretrained, WGISD: Wine Grape Instance Segmentation Dataset, and MKV2: Microsoft Kinect V2.