Nondestructive Banana Ripeness Classification using Neural Network

Non-Destructive Banana Ripeness Detection Using Shallow and Deep Learning: A Systematic Review

“…A single banana is called a finger, and a grouping of attached fingers of 10 or more bananas is called a hand [ 3 ]. In this review, out of the 35 studies, a total of 27 [ 3 , 4 , 5 , 6 , 9 , 10 , 11 , 17 , 18 , 19 , 20 , 21 , 22 , 23 , 24 , 25 , 26 , 27 , 28 , 29 , 30 , 31 , 32 , 33 , 34 , 35 , 36 ] have used the fingers of bananas, 3 [ 37 , 38 , 39 ] studies have used the fingers by hand and Chen et al [ 40 ] used a bunch of bananas. Mohamedon et al and Rodrigues et al [ 41 , 42 ] used a mixed sample of bananas, Mendoza and Aguilera [ 43 ] used the combination of fingers by hand and a single batch of bananas, and whereas Saragih and Emanuel [ 44 ] also used the used the combination of fingers and fingers and hand bananas.…”

“…Therefore, this section summarizes the relevant articles’ sensor types, the wavelength distance, and the color space distribution for the included studies. Out of the total considered studies (35), 18 [ 5 , 6 , 9 , 10 , 19 , 20 , 21 , 24 , 26 , 27 , 28 , 34 , 38 , 39 , 40 , 41 , 42 , 44 ] used a standalone digital camera. Hou et al [ 22 ] used spectrometers and Saadl et al [ 30 ] used a webcam.…”

“…Out of the 35 studies, only five [ 5 , 10 , 36 , 37 ] studies have provided the origin country information of the dataset, of which two [ 10 , 36 ] studies came from the Philippines, Altaf et al [ 37 ] were in Pakistan, and Mohapatra, Shanmugasundaram and Malmathanraj’s [ 5 ] study is in India. The rest of the studies have taken the data from commercial distributors, local markets, and self-created datasets but did not provide the origin country information, using existing online resources data and data from the ripening facility in Potsdam Germany, whereas 12 [ 11 , 19 , 21 , 27 , 28 , 29 , 30 , 31 , 34 , 38 , 39 , 41 ] studies did not report any information about the country origin of the dataset.…”

“…Medians were calculated from all the included augmentation and sensor studies for the quality assessment analysis. Out of all 30 quality assessment studies, 12 [ 6 , 9 , 10 , 17 , 18 , 19 , 20 , 25 , 35 , 38 , 41 , 44 ] have used 245 or more samples and 11 [ 4 , 5 , 22 , 23 , 27 , 30 , 33 , 37 , 40 , 43 ] studies have used values less than this amount. Among the four articles that have used a highly augmented dataset, Sabilla et al [ 21 ], as well as Marimuthu, Mohamed and Mansoor Roomi [ 39 ] have employed a median value of 2054 or more samples, and the other studies of Spaches [ 26 ] and Vaviya et al [ 34 ] used values less than this.…”

“…There are different stages used for the classification of banana ripeness. Among the 35 examined articles, 4 [ 5 , 22 , 36 , 43 ] examined seven stages, 4 [ 17 , 18 , 35 , 37 ] studies have used six stages, 6 [ 6 , 19 , 20 , 25 , 40 , 44 ] have used four stages, 12 [ 9 , 10 , 11 , 24 , 27 , 29 , 30 , 31 , 32 , 39 , 41 ] used three stages, and the remaining 3 [ 33 , 34 , 41 ] studies have used two stages. Lastly, Sabilla et al [ 21 ] used the seven categories of banana, and Iswari et al [ 28 ] employed three different types of fruits, namely, apple, banana and melon.…”