Recent Advancements in Fruit Detection and Classification Using Deep Learning Techniques

Table 2

Summary of the deep learning models applied to fruit classification.

Ref/year	DL model	Dataset	Dataset partition	Accuracy
[100] 2015	CNN	Personalized Dataset, UEC-FOOD100	—	80.8% SF, 60.9% MF
[101] 2017	Modified VGG	Personalized Dataset	80% Train, 20% Validation	95.6%
[102] 2017	MCNN	ImageNet	—	74% WDA 90% DA
[14] 2017	13-layer CNN	Veg Fruit Dataset	63000 Train, 1800 Test	94.94%
[103] 2018	PCNN + GAP	Fruit 360	80% Train, 20% Test	98.88%
[103] 2018	CNN FC-L	Fruit 360	80% Train, 20% Test	97.41%
[103] 2018	CNN FC-L	Dropout Fruit 360	80% Train, 20% Test	97.87%
[104] 2018	MAlexNet	Personalized ImageNet	80% Train, 20% Test	92.1%
[105] 2018	DCNN	Personalized	30082 Train, 7520 Validation, 6804 Test	90%
[74] 2018	6-layer CNN	Personalized	900 Train, 900 Test	91.44%
[106] 2018	8-layer CNN	VegFru	50% Train, 50% Validation, 50% Test	96.67%
[74] 2019	9-layer CNN	COCO apple class	70% Train, 15% Validation, 15% Test	99.78%
[107] 2019	LW models	TL, Fruit 360	80% Train, 20% Test	98.7%
[108] 2019	DCNN models	Fruit 360	80% Train, 20% Test	99.6%
[24] 2019	VGG-16 + GAP	SPD, Personalized	85% Train, 5% Validation, 15% Test	99.49%
[24] 2019	LA	SPD, Personalized	85% Train, 5% Validation, 15% Test	99.75%, 96.75
[39] 2019	M-GNet	Hyperspectral Images	2000 Train, 700 Validation, 125 Test	88.15% PRGB 85.93% LC 92.23% CK
[109] 2020	CAE-AND	Fruit 26, Fruit 15	85,260 Train, 38,952 Test	95.86%, 93.78%
[110] 2020	InterFruit	InterFruit	70% Train, 30% Test	92.74%
[111] 2020	VGGNet	—	—	—
[112] 2020	CNN SL	Orange Fruit	60% Train, 20% Validation, 20% Test	—
[109] 2020	ResNet-500	Fruit 26, Fruit 15	80% Train, 20% Test	93.59%, 91.44%
[109] 2020	DenseNet-169	Fruit 26 Fruit 15	80% Train, 20% Test	93.87%, 91.46%
[113] 2020	Deep CNN	Cheery	—	99.4%
[114] 2020	MobileNetv2	A O B	—	95% PB 93% WPB
[115] 2020	EDLS Fruits	Fresh, Fruit-360, Rotten for Classification	—	—

DL: deep learning, TL: transfer learning, SPD: Supermarket Produce Dataset, PT = pretrained dataset, PB: plastic bags, WPB: without plastic bags, A O B: apples, oranges, and bananas, WDA: without data augmentation, DA = data augmentation, SF: single food, MF: multi-food, PRGB: with pseudo-RGB images, LC: with linear combinations, and CK: with convolutional kernels.