Fish-Pak: Fish species dataset from Pakistan for visual features based classification

Fishes are most diverse group of vertebrates with more than 33000 species. These are identified based on several visual characters including their shape, color and head. It is difficult for the common people to directly identify the fish species found in the market. Classifying fish species from images based on visual characteristics using computer vision and machine learning techniques is an interesting problem for the researchers. However, the classifier's performance depends upon quality of image dataset on which it has been trained. An imagery dataset is needed to examine the classification and recognition algorithms. This article exhibits Fish-Pak: an image dataset of 6 different fish species, captured by a single camera from different pools located nearby the Head Qadirabad, Chenab River in Punjab, Pakistan. The dataset Fish-Pak are quite useful to compare various factors of classifiers such as learning rate, momentum and their impact on the overall performance. Convolutional Neural Network (CNN) is one of the most widely used architectures for image classification based on visual features. Six data classes i.e. Ctenopharyngodon idella (Grass carp), Cyprinus carpio (Common carp), Cirrhinus mrigala (Mori), Labeo rohita (Rohu), Hypophthalmichthys molitrix (Silver carp), and Catla (Thala), with a different number of images, have been included in the dataset. Fish species are captured by one camera to ensure the fair environment to all data. Fish-Pak is hosted by the Zoology Lab under the mutual affiliation of the Department of Computer Science and the Department of Zoology, University of Gujrat, Gujrat, Pakistan.

Fishes are most diverse group of vertebrates with more than 33000 species. These are identified based on several visual characters including their shape, color and head. It is difficult for the common people to directly identify the fish species found in the market. Classifying fish species from images based on visual characteristics using computer vision and machine learning techniques is an interesting problem for the researchers. However, the classifier's performance depends upon quality of image dataset on which it has been trained. An imagery dataset is needed to examine the classification and recognition algorithms. This article exhibits Fish-Pak: an image dataset of 6 different fish species, captured by a single camera from different pools located nearby the Head Qadirabad, Chenab River in Punjab, Pakistan.

Data
Classifying fish can be valuable for various purposes one of which is the identification of different fish species. Classifying fish accurately are beneficial for the study of fish diversity [1]. Aside from this, the grouping of fishes is additionally valuable for learning the deportment and interspecies cooperation of fishes in a typical natural condition [2]. In the field of machine learning and computer vision, classification of fish species from images is a multi-class recognition issue and is an attractive research domain [3]. Automated fish classification is very important in fisheries research as it helps in automated monitoring of fish species activities in the ponds, feeding and diseases behavior.  pixels and the resolution of 72 dpi. Fig. 1 shows the subset of head view images taken from the Fish-Pak randomly. Similarly Figs. 2 and 3 contains complete body view and scale view of 12 different instances. We have preprocessed the data and make each image background transparent. Detailed characteristics with respect to different fish features of Fish-Pak dataset are given in Table 1.

Camera specification and setting
A digital camera (Canon EOS 1300D) with a sensor type of CMOS bearing the resolution of 5202 Â 3465 (Mpix) and the sensor size of 14.9 Â 22.3 (mm) was utilized for all image collection. The mode of camera was Scene, with the selection of sub-category as Snow scene, as it demonstrated the best mode for the unusual light condition of the case; with 14 megapixels picture measure (5184 Â 3456 pixels) in 3:2 extents, glimmer and face discovery deactivated in the case when capturing fish body and scale. Furthermore, 2.5 instants zoom to all the more likely spot the fish head on the picture and less misuse of the image area with foundation. RGB shading space is selected for each of the images in JPG format, 8 pixels for each shading layer, adding 256 shades for every RGB layer.

Deep representation of feature maps
We have applied CNN on the body images of Fish-Pak dataset and extract deep feature maps that can be found in Fig. 4. VGGNet [4] was selected to obtain the internal representation of the feature map from the 2nd convolutional layer with the first 64 maps. The kernel size for the experimentation was 3 Â 3 with the pad of (1, 1) and stride of 128. We can see from Fig. 4, the more layer becomes deeper the more image becomes not interpretable by humans.  Mouth is inferior.
Mouth is upturned.
Mouth is wide and slightly superior.
Mouth is large and slightly oblique.
Mouth is terminal to sub terminal. Head is equilateral.
Head is isosceles.
Head is broad.
Head is large and broad. e Head is compressed and slightly pointed. Snout is depressed and projects beyond the jaws.

Snout is blunt.
Snout is bluntly rounded. Snout is short and blunt. Snout is long and blunt.
Snout is very short. Lips are non-fleshy and firm

Fin rays
Dorsal fin has 12e13 fin rays.
e e e Pectoral fin has 14 e18 soft rays.
Pectoral fin has 15e20 rays. Pelvic fin has 9 rays. e e e Pelvic fin has 8 or 9 soft rays.
e Anal fin has 7 rays. e Anal fin has 8 rays.
Anal fin has 12 rays. Anal fin has 4e6 soft rays.
Anal fin has 8e10 rays.
Caudal fin has 19 rays. e e Caudal fin has 21e22 rays. Caudal fin has 19 soft rays