Publiscience Vol. 4 Issue 1 How to cite this article: CSE: Soldevilla TP, Redaniel DNAA, Sy LCV, Nulla MMA. 2021. Using a sssNet Convolutional Neural Network (CNN) with Support Vector Machine (SVM) algorithm to identify formalin presence in images of eyes of Chanos chanos. Publiscience. 4(1): 99–104. APA: Soldevilla, T.P., Redaniel, D.N.A.A., Sy L.C.V., & Nulla, M.M.A. (2021). Using a sssNet Convolutional Neural Network (CNN) with Support Vector Machine (SVM) algorithm to identify formalin presence in images of eyes of Chanos chanos. Publiscience, 4(1), 99–104. For supplementary data, contact: publiscience@wvc.pshs.edu.ph. Using a sssNet Convolutional Neural Network (CNN) with Support Vector Machine (SVM) algorithm to identify formalin presence in images of eyes of Chanos chanos (milkfish) THOREENZ P. SOLDEVILLA, DOMINIC NATHANIEL ANTHONY A. REDANIEL, LANCE CHRISTIAN V. SY, and MARIA MILAGROSA A. NULLA Philippine Science High School Western Visayas Campus - Department of Science and Technology (DOST-PSHSWVC), Brgy. Bito-on, Jaro, Iloilo City 5000, Philippines Article Info Abstract Using image processing, eye turbidity of formalin-treated Chanos chanos (milkfish) was statistically proven to be significantly different from those untreated. However, automation of such processes was yet to be explored. This study aims to use a sssNet Convolutional Neural Network (CNN) with Support Vector Machine (SVM) algorithm to identify formalin presence in milkfish. Ninety percent of 420 formalin-treated milkfish images and 420 untreated images, each with an indicated day of image capture, were subjected to feature extraction and classification using sssNet-SVM. The remaining 10% of the dataset was used to validate the algorithm’s performance. The algorithm garnered 98.16 to 99.15% validation accuracy for identifying formalin presence. However, seven-day feature map analysis reveals that the algorithm struggles to determine formalin presence in treated samples using their images that were captured one or two days after the samples’ dousing in formalin. Submitted: Apr 30, 2021 Approved: Jun 21, 2021 Published: Aug 30, 2021 Keywords: Chanos chanos formalin image processing neural network SVM Introduction. - The Philippines is a fish- producing country that ranks 11th in global fishing production [1]. The main aquacultural produce of the country, Chanos chanos, locally known as bangus or milkfish, accounts for 2.4% of the national fisheries production [2]. Fish are highly perishable food, with storage times for tropical species ranging from 6-40 days [3]. Due to this limitation, various preservation techniques have been devised to prolong its freshness in order for such to be marketable for longer periods of time. One of the chemicals used to preserve fish is formalin [4,5], a solution consisting of 37% formaldehyde, a known respiratory disease enabler [6]. Formalin can also cause early protein denaturation which compromises fish quality [7]. Several studies used chemical analysis methods to detect the early presence of formalin in meat and fish such as spectrophotometry [8] and formalin rapid testing [9]. However, methods regarding chemical analysis are labor-intensive and time-consuming, while rapid test kits are not readily available in the market. This limitation was addressed by Cadorna et al. [8]. that used computer vision techniques such as image processing to detect formalin presence in milkfish by evaluating its eye, a method similar to most computer vision techniques that measure fish freshness. Image analysis has implied that the eye of formalin-treated fish became cloudy after a seven-day period as opposed to untreated fish which almost retained its appearance. The study then quantified the eye turbidity by capturing the image of the fish, splitting the channels into HSV (Hue, Saturation, and Value) components, and determining the intensity of each color space using an image processing tool. The study then found out that with values below 0.05 level of significance, the value components of the eye images +-of formalin treated and untreated C. chanos are significantly different. Such has opened the possibility of utilizing eye turbidity to be used for automation of formalin detection using imagery. Since automation of fish quality [11] and classification of eye appearance [12] is possible, formalin detection in C. chanos can be done using supervised machine learning. Algorithms for classifying fish samples according to their quality are using two distinct methods — feature extraction and classification. These are done by first enhancing the images using methods such as blob extraction and border smoothing [11], as well as eye masking [12]. Then, the processed images are loaded into algorithms such as k-Nearest Neighbor, Support Vector Machine (SVM), or Feed-forward Artificial Neural Network (ANN). Hence, it is implied that feature extraction and classification algorithm methods for machine learning are always interdependent of each other. However, feature extraction is a tedious process that requires enhancing images manually before being analyzed or loaded into an algorithm. To be able to overcome this limitation, an algorithm that unifies feature extraction and classification shall be utilized.