WO2017221259A1

WO2017221259A1 - Automatic recognition of indian prawn species

Info

Publication number: WO2017221259A1
Application number: PCT/IN2016/000235
Authority: WO
Inventors: Jyothi S.; Mamatha D.M.; Nagalakshmi G.; Himabindu K.
Original assignee: S Jyothi; D M Mamatha; G Nagalakshmi; K Himabindu
Priority date: 2016-06-23
Filing date: 2016-09-30
Publication date: 2017-12-28

Abstract

The present invention relates to an automatic prawn species recognition system by novel pattern recognition technique to identify prawn species. The technique is based on a combination between robust feature selection, image segmentation and geometrical parameter techniques. The system comprises of following modules based on digital Image Processing Techniques and Neural Networks technique. The system involves firstly harvesting prawns from fields with different sizes of adult prawn and fedding through a computer vision system in any orientation on a conveyor belt. Secondly capturing image of input prawn in which back ground should be pure white for easy pre-processing of the image. Thirdly image pre-processing of the captured image which comprising of resizing, converting into gray scale images, removing of noise, normalization for the noise removal image, and extracting object from white background. Fourthly image segmentation by segmenting the extracted object into individual blocks and extracting necessary segment blocks. Fifthly features extraction from the segment blocks by block merging based on feature vectors and matching with histogram based comparison stored in an existing database for feature training. Finally classification of prawn images from the extracted features using Neural Networks built by taking the features from existing database.

Description

TITLE: AUTOMATIC RECOGNITION OF INDIAN PRAWN SPECIES

FIELD OF THE INVENTION

The present invention relates automatic prawn species recognition system. More particularly the present invention relates to a Pattern recognition to classify the prawn species with neural networks using external characteristics of the prawns.

BACKGROUND OF THE INVENTION AND PRIOR ART:

Food authenticity is an issue of major concern for food authorities, since incorrect food labeling represents commercial fraud to the consumers, in particular when this implies the replacement of one ingredient by another of lower commercial value (Lockley & Bardsley, 2000). The fish products sector is characterized by the high commercial value of many of its products, this being specially relevant in the case of prawns, when they are sold as whole specimens and when they are included as high- value ingredients in pre-cooked food products. Among prawn species, morphological characters are particularly used for species to differentiate their phenotypic differentiations.

Several efforts have been devoted to the recognition of digital image but so far it is still an unresolved problem, due to distortion, noise, segmentation errors, overlap, and occlusion of objects in color images. Recognition and classification as a technique gained a lot of attention in the last years wherever many scientists utilize these techniques in order to enhance the scientific fields. Prawn recognition and classification still active area in the fisheries domain and considered as a potential research in utilizing the existing technology for encouraging and pushing the fisheries researches ahead. Recognition ability from image can also be applied into computer system for automated recognition based on not just the text input but also the shape of images. The recognition of patterns (prawns) from aquired images of documents has been a problem that has received much attention in the fields of image processing, pattern recognition and artificial intelligence.

Neural Networks are capable of detecting patterns in their input and partitioning those patterns into categories without requiring that the number or types of categories to be predefined. The inputs for the Neural Networks are physical features of species extracted from feature extraction techniques and general characteristics of species. The networks performed well as measured by (1) the average correlation between the input vectors and the weight vectors for each category, and (2) the ability of the networks to classify novel species.

Ricardo Gutierrez-Osuna, Texas University [Ric], explained the use of classification of prawn species in the real world by identifying the incoming prawn species and then separate Penaeus monodon from Penaeus indicus automatically (Figure 1).

Mutasem Khalil Alsmadi et al., [UM13] have proposed a system to find an isolated pattern of interest of image based on the extraction of robust features. A fish Image has been taken and then image segmentation is done based on colour texture measurement. The system has been applied to 20 families of fish. For the extraction of colour and texture, Gray level co-occurrence Matrix method has been chosen. The features of the image are extracted based on its size and shape measurements using local geometric approach which uses angles and distance measurement from anchor or landmark point locations. White DJ, Svellingen et al., [WSS06] described methods for Sorting and identification of a species of fish (i.e flat fish or round fish)using computer vision machine when the fish is moving on a conveyor belt underneath a digital camera. The programming techniques were used to measure length and identify species. Sebastien et al., [SFF00] presented an automated system for counting the fish by species. This system is mainly used in fish ways for surveying and monitoring of the fishes. Images are being sent to computer for analysis, classification and counting. Analysis is done by using features like Moments Invariants and Fourier boundary descriptor

Boaz Zion et al., [BVV07] developed a method for Real time under water sorting of edible fish species. Some of the fishes namely Common crap, St.Peter's fish and Grey mullet were sorted based on the species when they are swimming in pond water which contains algae, when the fishes are swimming their images were acquired by computer vision system. A relative constant distance from the camera was maintained. The features like size and orientation are extracted by using an algorithm. Usama et al., [UM13] presented a method to optimize performance of Back-Propagation classifier for the classification of the fish by using the Hybrid Memetic algorithm. The features are extracted based on colour signatures of all fish images by using Histogram technique and Gray level co-occurrence Matrix methods. Murat O.Balaban et al., [MGBD11] have adopted Machine vision in processing of sea food for performing many tasks given Figure 3.

M.S.Nery et al., [Msn05] developed a new fish classification technique based on feature selection. Classifier uses a general set of features of fish like size, shape, colour signatures and texture measurements as a priori information. The colour signatures are related to the dorsum and ventral colours of the fish. N.J.C.Strachan [SNA90] has developed a fish species recognition method based on the shape analysis of images which is done using a database of shapes drawn from several photos of seven different types of fishes on the computer. Three approaches were used for this they are invariant moments, shape descriptors, optimization of mismatch.

G.T.Shrivakshan et al., [SCI 1] proposed a method for classification based on image processing using wavelet transformation for identifying the edge specifically two dimensional Haar wavelet transformation is used. Eiji Morimoto et al., [EYM09] performed a research for the identification of species of fish because automated selection of species of fish is essential to the markets. A quantification method of characteristics of the fish image was developed which is used for the identification of fish species based on the quantified characteristics. The identification method used is Neural Networks which learn to differentiate between various fish species using reference points. Reference points are characteristics that are extracted from images.

Frank et al., [FB01] describes a system to recognize species of fish by using computer vision and neural network. The computer vision system measures the number of features of fish when seen in camera which is perpendicular to conveyor belt. The features extracted are the heights and widths at different projected areas of the fish. These features are used as input values to the neural network. The network is trained to identify the species of the fish from the given input data. Sergio et al., [SB07] identified the fish age using analysis and classification with kernel methods. The knowledge of age in fish population is very much important in the stock assessment. The scientist of fisheries conventionally used fish otoliths i.e., the classified structure of the inner year which is used for finding out the age of the fish because of the shape changes during the life time of the fish. Strachan at el., [Str94] have studied and tested a prototype system at the sea for sorting the fish based on species and size. Fishes were placed on the conveyor belt and grabbed their images. Using the fish length to width ratio the round and flat fish were differentiated. Strachan et al., [Str93] described the algorithms used for generating the shape and colour descriptors of the fish images by which fish species can be recognized. The descriptors are not affected by the deformations and bending of the fish. Various studies performed on computer vision techniques are compared as shown in table 1.

Table 1: Comparison of various Computer Vision Techniques

Eii Morimoto et identification

Reference points Neural Networks N/A al. of fish species

Heights and

Frank et al. fish species Neural Networks 95%

widths

Sergio et al. fish age Shape PCA & SVM N/A

Sorting of fish Length to width

Strachan et al. Shape and size 99%

species ratio

identification

Strachan et al. Shape and colour 98-100%

of fish species

Common carp -91%

Algorithm based on St-Peters

B.Zion et al. Sorting of fish Geometric

moment invariants fish-91%

grey mullet- 100%

Length

N.J.C Strachan Geometric Neural Networks N/A

measurement

E.Misimi et al. Sorting of fish Colour Segmentation N/A

Ching-Lu Hsieh Fish Length Projective &Hough

Geometric N/A et al. measurement transformation

Removal of Morphological Thresholding

Linp PP et al. N/A

shrimp Heads features segmentation

Shrimp

Length,area,pixel Thresholding

Harbitz carapace N/A

and weight segmentation

length

Weight of WAP Model and

Width,area and

Peng-min shelled segmentation 96.9%

perimeter

shrimp

Poonpat Weight of No of pixels of

Regression analysis N/A

Poonnoy et al. sushi shrimp projected area

Shrimp

Linear Regression,

Mohebbi et al. dehydration Colour N/A

Neural Networks

level

Quantification

Zayde alcicek et

of colour of Colour Statistical Methods N/A al.

tiger prawns

Luzuriaga et al. white shrimp Count and colour Regression analysis N/A

Identification

Color, Shape and

V. Sucharitha of three Neural Networks 85%

Texture

prawn species

Sucharitha [Sue 14] described about classification of penaeid prawn species based on colour, shape and texture using neural networks in her thesis. But by using color extraction there is no complete accuracy for extractions because the color and shape of a prawn may not be static, color and shape changes occur depends on the increasing of the life span. It is not described about the segmentation of the prawn.

In the previous work, contours and lengths are used in automatic fish recognition system because of simple structure of fish. Since prawns have delicate legs and antenna and Prawns have complex structure compare to fish, contour (direct shape) extraction is not sufficient to recognize prawn.

All the above methods are used for specific regions with specific species only and also most research was done in Asian pacific countries. Aquaculture in India is also growing field even in food processing and industrial food technology.

One of the most remarkable developments in the fisheries of India in recent years has been the phenomenal expansion in the prawn fishery. The main factors responsible for the development of the prawn fishery were the discovery of large resources of prawns on this coast through exploratory fishing conducted by the State Fisheries Department, the Central Institute of Fisheries Technology, and the Offshore Fishing Station of the Government of India in collaboration with the Central Marine Fisheries Research Institute; introduction of modern processing techniques like freezing and canning; and ever-increasing demands for prawns from external markets.

By the increase in the demand for seafood marine aquaculture has become one of the leading growing industries worldwide. The visual attributes are the most important parameter for foods in general and specifically in aquatic foods. The consumer takes decision regarding purchasing based on the price and product defined by the visual attributes like the species of the prawn. The sorting of the prawn species manually is carried out on all commercial and research fishing vessels. This process is very slow and is the rnajor drawback in terms of efficiency and requires increased labor. At present all the prawns caught are manually graded by species and also weight. So there is a requirement of automatic sorting system of prawns which is capable of recording the data of the species. This requirement is driven by the need to reduce the labor. In general it is not possible practically for the normal human beings to examine and recognize the large amount of prawn species and it is extremely cost. Hence automatic prawn species recognition system is required to overcome all the errors caused by conventional prawn species classification methods which is based completely on the human expertise. The computer based system must be used to replace the human subjective evaluation for the successive evaluation of various sea foods. It would be very helpful to carry out the behavioral study on prawns for the automatic classification of prawns, because each of the prawn species has got its own distinctive patterns. Automatic prawn species recognition has not been well established being the fact that a lack of research in this field and the complexity in getting the database. Prawn species recognition is one of the major concerns at present and there is a huge requirement for several researches to deal with better prawn species recognition system. Thus there exists a need in the state of art to develop an automatic prawn species recognition system focusing on Pattern recognition to classify the prawn species with neural networks using external characteristics of the prawns.

References

[BVVI07] Boaz Zion, Victor Alchanatis, Viacheslav Ostrovsky, Assaf Bark arid

Iian Karplus. (2007), "Real time underwater sorting of edible fish species", Computer and Electronics in Agriculture. Vol 56, Issue 1, pages 34-55

[BYB94] Balaban, M.O. Yeralan, S. and Bergmann, Y. (1994), "Determination of count and uniformity ratio of shrimp by machine vision". Journal of Aquatic Food Product Technology, 3:43-58

[EYM09] Eiji Morimoto, Yuichiro Taira and Makota Nakamura. (2009),

"Identification of Fish Species using Neural Networks", Journal of National Fisheries University 58(1) 65-71.

[FB01] Frank Stobeck and Berent Daan. (2001), "Fish species recognition using computer vision and a Neural Network", Fisheries Research Vol 51, issue 1, pages 11-15.

[LB00] Strachan N.J.C, Nesvadba P. and Allen. A.R. (1990), "Fish species recognition by shape analysis of images", Pattern Recognition, 23(5):539-544.

[LB08] Lockley A.K, & Bardsley,R.G, (2000), DNA-based methods for food identification, Trends in Food Science and Technology, 11, 67-77. Bai X., Yang X. and Latecki J.L. (2008), Detection and Recognition of Contour Parts Based on Shape Similarity.Philadelphia, USA,

[MGBD11] Murat O.Balaban, Gulgun F.Unal Sengor, Bahar Gumus, Dilsat

Cirban. (2011), "Image analysis to estimate the weight of fish", ICEFl l.

[Ner09] Nery M.S. (2009) "Determining the appropriate feature set for fish classification tasks", in Computer Graphics and Image Processing Brasil, Doi: 10.11.09/SIBGRADI

[Ric] Ricardo Gutierrez-Osuna, http://research.cs.tamu.edu/, Texas

University.

[SB07] Sergio Bermejo, Bras Monegal. (2007), "Fish age analysis and classification with Kernel methods". Pattern Recognition Letters 28(10):1164-1171.

[S.C11] Shrivakshan G.T, Chandrasekhar C. (2011) "Shark Fish

Classification through image processing using wavelet transformation and enhanced edge detection technology", International journal of comp. Tech. Vol2(4), 773-783.

[SFF00] Sebastien cadieux, Francois lalonde and Franclois Michand. (2000),

"Intelligent system for Automated Fish sorting and counting". Doi: 10.1 109/ IROS, 893195.

[SNA90] Strachan NJ.C, Nesvadba P. and Allen. A.R. (1990). "Fish species recognition by shape analysis of images", Pattern Recognition, 23(5):539-544.

[SNA90] Strachan NJ.C, Nesvadba P. and Allen. A.R. (1990), "Fish species recognition by shape analysis of images", Pattern Recognition, 23(5):539-544.

[Str93] Strachan NJ.C. (1993), "Recognition of fish species by colour and shape". Image Vision Comp. 11(1):2-10.

[Str94] Strachan NJ.C. (1994), "Sea trials of a computer vision based fish species sorting and size grading machine mechantronics", 4(8):773- 783. [Sue 14] Sucharitha V. (2014), Ph. D. Thesis, "Classification of Penaeid Prawn Species based on Colour, Shape and Texture using Neural Networks", SPMVV, Tirupati.

[UM13] Usama A.Badawi, Mutasem Khalil Sari Alsmadi. (2013), "A Hybrid memetic algorithm (Genetic algorithm and Great Delugu Local search) with Back-propagation classifier for Fish Recognition". IJCSI, Vol 10, Issue 2, Nol, March 2013.

OBJECT OF THE INVENTION:

The main objective of the present invention is to develop a novel prawn species recognition system using Neural Networks to identify prawn species automatically without specialized taxonomists.

Another objective of the present invention is to employ hardware and programming techniques to identify prawn species in real time.

Yet another objective of the present invention is to recognize prawn species by feding Prawns through the computer vision system in any orientation on the conveyor belt and species are determined based on the feature extraction.

Yet another objective of the present invention is to develop a prawn recognition system, which has reduced cost of human labor and time, increased efficiency and throughput.

Further object of the present invention is to utilize the developed prawn species recognition system for identification of prawn species

BRIEF DESCRIPTION OF DRAWINGS:

Figure 1 depicts Automatic Classification of Monodon and Indicus. Figure 2 depicts Stages in automatic prawn identification system. Figure 3 depicts Block diagram of the fish classification system. Figure 4 depicts Block diagram of the fish classification system. Figure 5 depicts Prawn's images acquired from harbors and hatcheries. Figure 6 depicts a) Original Image b) Gray Scale output images for original conversion c) Modified conversion.

Figure 7 depicts Object Extraction of Prawn Image.

Figure 8 depicts Block division and database storing for a given input prawn Figure 9 depicts Histogram images for the block images

Figure 10 depicts (a) Result after removing unwanted block using tinning process; (b) Result after removing legs based on pixel positions; (c) Result after merging block together.

Figure 11 depicts Feature Extractions based on pixel positions of a Prawn

Figure 12 depicts Graphical User Interface of feature extraction

Figure 13 depicts Proposed Multi Class Backpropagation Neural Network

Figure 14 depicts Comparison and Recognition Rate for the different prawns

Figure 15 depicts Comparison proposed multi class BPNN with existing methods

SUMMARY OF THE INVENTION:

DETAILED DESCRIPTION OF THE INVENTION:

The present invention relates to an automatic prawn species recognition system by novel pattern recognition technique to identify prawn species.

Pattern recognition is a novel prawn classification methodology based on a combination between robust feature selection, image segmentation and geometrical parameter techniques using artificial neural networks. Unlike existing works for prawn classification, which propose descriptors and do not analyze their individual impacts in the whole classification task and do not make the combination between the feature selection, image segmentation and geometrical parameter, this study proposes a general set of features extraction using robust feature selection, image segmentation and geometrical parameter and their correspondent weights that should be used as a priori information by the classifier. The classification problem involved the identification of image prawns; family, Scientific Name, English name and local name.

The images of different prawns are isolated from one another and from the background. The physical differences between the types of prawns are length, lightness, width, number and shape of fins, position of the mouth, and so on. The information from a single prawn is then sent to a feature extractor, whose purpose is to reduce the data by measuring certain features or properties. These features are then passed to a classifier that evaluates the evidence presented and makes a final decision as to the species.

The automatic recognition of prawn species by their patterns is a very interesting field. Due to the specific properties of prawn species, it is necessary to develop new image processing techniques. The main idea of this invention is to study, how different patterns of prawn structure can be recognized and classified through neural networks.

Although more than 250 species of Penaeids and an equal number of Non-Penaeids have been recorded in the landings, only a few of them contribute to the commercial prawn fishery of the area. The species of commercial importance are three major families, namely Penaeidae, Sergestidae and Palaemonidae. Some of the sub species of the major families are Penaeus monodon, Penaeus indicus, Penaeus Vennamei, Metapanaeus monoceros, Metapenaeus affinis, Metapenaeus dobsoni, Metapenaeus brevicornis, Parapenaeopsis stylifera, Solenocera indica, Acetes indicus; Acetes erythraeus; Leander termipes, Leander styliferus.

First the automatic prawn recognition work was done by the inventors based on color, shape and texture. In the previous work, it was observed that color will be changed from time to time, legs and antenna will be broken. Therefore, presently complex features were extracted and then considerd prawn shape without legs and antenna. Existing image processing techniques are not directly implemented to extract prawn features like rostrum, carapace, abdomen and telson. The proposed invention, ARIPS, developed many new algorithms to extract the features of prawn images for automatic recognition of prawn species. For this invention, prawn images collected from fields with different sizes of adult prawn.

The proposed system has the internal architecture of as given in Figure 2. This architecture contains two main parts one is feature extraction which is based on digital Image Processing Techniques and other is Neural Networks technique.

The main aim of the system is to perform an automatic classification of prawn species for a new pattern based on the existing data by using Neural Network technique. To achieve this Neural Network (A Back propagation Multi Class Neural Network Model) is built by taking the features from the existing database. Then the network is trained by using the selected training method and learning rule. The network is trained until it reaches its desired output. Once the network is trained then it can be used to perform the classification automatically for a new prawn after. This architecture is implemented in the proposed invention.

The proposed system is divided into five modules. The first module is the capturing of the input prawn by using DSLR camera in the Left direction i.e head section is in left position of the capturing mode and back ground should be pure white so that it is easy to pre-processing the image in the next module The second module is Image pre-processing in which all the images are resized, converted into gray scale images, Removing of noise, Normalization is done for the noise removal image, and finally extracting the object form white background.

The Third module is the image segmentation, segmenting the input prawn image into individual blocks and extracting only wanted segment blocks by using divide and conquer methods. After that features are extracted from the given input prawn image. The objective of this module is to find the edges and contour in a given image using sobel edge detection technique. All the images that are matching with respect to the Histogram based comparison are stored in database for feature training. Histogram based comparison is done and matching images are shortlisted based on Euclidean distance. After segmentation we extract each block based on the feature vectors. In our proposed system we consider Rostrum, Carapace, Abdomen and Telson are the major features for feature extraction.

The Fourth module is the classification of prawn images using Neural Networks. The objective of this module is the extracted prawn features using Digital Image processing techniques are given as input to the neural network which is working on Back propagation algorithm to perform classification. The resultant value of this module determines whether a prawn belongs to particular species or not.

The first stage of any computer vision system is said to be the image acquisition stage. After the image has been obtained, different methods of processing can be applied to the image to perform different vision tasks required. However, if the image has not been acquired in a required manner then the intended tasks may not be achievable, even image enhancement is done. The detailed description of proposed invention is given in figure 4.

Among many types of image formats like png, tiff, bmp, jpeg or jpg, jpeg format of the image was choosed because jpeg format takes less memory space, easily viewable from internet, use millions of colours and perfect for many types of images. The different species of Prawn images that are acquired from harbors, hatcheries. For training and testing for the prawn identification more than 1000 prawns are captured on which some of them are shown in figure 5.

For the extraction of Prawn features, all the colour prawn images were converted into gray scale. When converting an RGB image to grayscale, RGB values were taken for each pixel and make as output a single value reflecting the brightness of that pixel. One such approach is to take the average of the contribution from each channel:

(R+B+C)/3.

However, since the perceived brightness is often dominated by the green component, a different, more "human-oriented", method is to take a weighted average. The original images of three species of prawns are and the gray scale images are shown in figure 6. Noise removal and normalization is performed on gray scale prawn image. The source of noise in digital images arises during image acquisition or during image transmission. In the present study Gaussian noise is considered.

Next object extraction method is proposed to differentiate foreground and background image frame based on morphological operations. Morphology is a technique of image processing based on shape and form of objects. Denoting an image by f(x) and the structuring function by b(x), the grayscale dilation of /by b is given by:

(/ Qb) = sup_yeE[f(y) + b(x - y)] --(1) where "sup" denotes the supremum. Similarly, the erosion of /by b is given by

(/ 0b) = M_yeE[f(y) - b(y - x)] --(2)

Advanced image processing and image enhancement tools have been used for maximum accuracy of the results and to identify the particles accurately from the image without even missing a single object. Figure 7 is shown the object extraction of the prawn image. Then resizing of the object is required for feature extraction. The resizing of an image returns image B that is scale times the size of A. This is very much essential to consider custom size of the prawn.

In Image segmentation process, initially prawn image is divided into 3 x 3 blocks, for each block Discrete Wavelet Transformation (DWT) is applied. To portion Image the framework first partitions the given input prawn Image into 3x3 little squares relies on upon execution time. Second request minute is considered as highlight vector of wavelet coefficients of high recurrence bands. To get these minutes daubechies-4 wavelet change is utilized. The daubechies wavelets, are based on the effort of Ingrid daubechies, which are born from the family of orthogonal wavelets defining a distinct wavelet change and characterize by a maximal amount of available moments for some given support. These characterizing maximal amount of moments are based on the wavelet and scaling functions of daubechies. There are four wavelet and scaling function coefficients of daubechies D4 transform which are as follows:

So = (l+ 3)/(4V2)

Si = ^'(3+V3)/(4V2)

5₂ = (3-V3)/(4V2)

5₃ = (1-V3)/(4V2)

Each step of the wavelet transform applies the scaling function to the data input. If the unique data set has N ideals, the scaling function will be practical in the wavelet transform step to compute N/2 smoothed values. In the structured wavelet transform the smoothed values are stored in the lower half of the N element input vector. Than the final wavelet function coefficient values are as follows:

F₀ = S₃

F,= -S₂

F₂ = S,

F₃ = -So

By taking the inner product of the coefficients the scaling and wavelet function are calculated for the four data values. The equations of the four data values are shown below:

Scaling functions for daubechies D4:

DrS₀W₂i+S₁W_2i+1+S₃W_2i+3

Dp] =S₀W[2i]+SiW[2i+l]+S₂W[2i+2]+S₃W[2i+3];

Wavelet functions for daubechies D4:

A_i = F₀W_2i+F₁W_2i+1+F₃W_2i+3

A[i] = F₀W[2i]+F₁W[2i+l]+F₂W[2i+2]+F₃W[2i+3];

Each iteration in the wavelet transform step calculates a scaling function value and a wavelet function value. The index /is incremented by two with each iteration, and new scaling and wavelet function values are calculated. Snippets of wavelet coefficients in different recurrence groups have been appeared to be successful. The block divisions and process are shown in figure 8. The figure 9 shows the different histograms for the block division which are generated by the given input prawn for removing the unnecessary blocks. After considering the main blocks using thinning process, first legs and antennas are removed by proposed pixel position & intensity algorithm. Then finally blocks are merged based on region adjacent graph (RAG) to get body of the prawn. Figure 10 gives outputs of the above process.

Digital images result from hit-or-miss transformations in which a structuring element of dots is ANDed with a scene to produce an image divided into pixels. Structuring elements consisting of matrices are then convolved with the image to accomplish image processing and measurement. Based on the dimensionality variations of pixel location the initial and final position of structuring element of a rostrum can be found. To measure the length of carapace an accurate slide calliper was used. Segmentation was done to each shrimp from its background using intensity thresholding. A precision of 0.43 mm was yielded by a linear model on a log-log scale of length in relation to pixel area. Using image analysis the time spent was less than 0.01s per prawns. Prawn abdomen is extracted based on continuous sequence of pixel coordinators of foreground image in x-direction until the pixel coordinators of foreground image changes in y-direction. Prawn telson is extracted based on continuous increasing sequence of pixel coordinators of foreground image in x- direction and continuous decreasing sequence of pixel coordinators of foreground image in x-direction. Figure 11 describes the Extractions of Features based on pixel positions of a Prawn. Graphical User Interface (GUI) design is given Figure 12.

Neural networks are formed from trillions of neurons (nerve cells) exchanging brief electrical pulses called action potentials. Computer algorithms that mimic these biological structures are formally called artificial neural networks to distinguish them from the squishy things inside of animals. Neural network research is motivated by two desires: to obtain a better understanding of the human brain, and to develop computers that can deal with abstract and poorly defined problems.

A neural network is a set of interconnected layers, in which the inputs lead to outputs by a series of weighted edges and nodes. The weights on the edges are learned when training the neural network on the input data. The direction of the graph proceeds from the inputs through the hidden layer, with all nodes of the graph connected by the weighted edges to nodes in the next layer. To compute the output of the network for any given input, a value is calculated for each node in the hidden layers and in the output layer. For each node, the value is set by calculating the weighted sum of the values of the nodes in the previous layer and applying an activation function to that weighted sum.

The Backpropagation neural network (BPNN) is a multilayered, feed forward neural network and is by far the most extensively used. It is also considered one of the simplest and most general methods used for supervised training of multilayered neural networks. Backpropagation works by approximating the non-linear relationship between the input and the output by adjusting the weight values internally. It can further be generalized for the input that is not included in the training patterns (predictive abilities). The Backpropagation neural network has two stages, training and testing. During the training phase, the network is "built" with sample inputs. During the testing phase, the network is "shown" sample inputs and the correct classifications.

For multi class classification, the typical approach is to have n output neurons in the final layer. They represent the different classes. In the end, the neuron which has the highest prediction 'wins' and that class is predicted. The following figure 13 shows the topology of the proposed multi class backpropagation neural network that includes 5-input layers, hidden layers and n- output layers. In our proposed multi class BPNN, dynamically the new and unknown species can be added at any time.

The Multi class network is trained to for Penaeus Indicus, Penaeus mergniensis, Penaeus Monodon, Metapenaeus dobsoni , Penaeus semisulcatus and so on. 250 samples of vennamei, 250 samples of monodon and 250 samples of Indicus are given as test images. The algorithms are coded and tested in MATLAB. The differences observed is with respect to Rostrum, Carapace, Abdomen , Telson and body values. The threshold is 0.3 on either side. Out of 250 images of Penaeus Indicus only 248 are identified correctly and accuracy is 98%. Out of 250 images of Penaeus Monodon only 248 are identified correctly and also accuracy is 98%. With these limited range inputs successfully, the network has been trained with higher rate of accuracy. Therefore, the overall accuracy of prawn species classification is 98% as shown in table 2. Table 2: Recognition Rate of some of the tested prawns

In one of the preferred embodiment the present invention relates to an automatic prawn species recognition system by novel pattern recognition technique to identify prawn species. The technique is based on a combination between robust feature selection, image segmentation and geometrical parameter techniques. The system reduces the cost of human labor, and time, and further increases efficiency and throughput without the need of specialized taxonomist. The system comprises of following modules based on digital Image Processing Techniques and Neural Networks technique. The system involves firstly harvesting prawns from fields with different sizes of adult prawn and fedding through a computer vision system in any orientation on a conveyor belt. Secondly capturing image of input prawn in which back ground should be pure white for easy pre-processing of the image. Thirdly image pre-processing of the captured image which comprising of resizing, converting into gray scale images, removing of noise, normalization for the noise removal image, and extracting object from white background. Fourthly image segmentation by segmenting the extracted object into individual blocks and extracting necessary segment blocks. Fifthly features extraction from the segment blocks by block merging based on feature vectors and matching with histogram based comparison stored in an existing database for feature training. Finally classification of prawn images from the extracted features using Neural Networks built by taking the features from existing database.

As per the invention in the automatic prawn species recognition system the features includes Rostrum, Carapace, Abdomen and Telson.

In accordance with the invention a modified gradient noise removal algorithm is employed to detect accurate features of Prawn images in the recognition system.

According to the invention an extraction algorithm is employed to extract foreground images from background in the recognition system.

As per the invention divide and conquer approach is developed for the block division, identification, modification and merging in the recognition system.

In accordance with the invention the feature extraction algorithms are generated to extract rostrum, carapace, abdomen and telson from the merged structure of a Prawn.

According to the system multi-class back propagation neural network model is employed to recognize the prawn species

As per the invention the classification involves the identification of image prawns; family, Scientific Name, English name and local name.

From the foregoing it will be appreciated that, although specific embodiments of the invention have been described herein for purposes of illustration, various modifications may be made without deviating from the spirit and scope of the invention. Accordingly, the invention is not limited except as by the appended claims.

Claims

WE CLAIM,

1. An automatic prawn species recognition system by novel pattern recognition technique based on a combination between robust feature selection, image segmentation and geometrical parameter techniques to identify prawn species thereby reducing the cost of human labor, and time, and further increases efficiency and throughput without the need of specialized taxonomist, wherein the claimed system comprises of following modules based on digital Image Processing Techniques and Neural Networks technique:

a. harvesting prawns from fields with different sizes of adult prawn and fedding through a computer vision system in any orientation on a conveyor belt;

b. capturing image of input prawn wherein back ground should be pure white for easy pre-processing of the said image ;

c. Image pre-processing of the captured image of step (b) comprising of resizing, converting into gray scale images, removing of noise, normalization for the noise removal image, and extracting object from white background;

d. image segmentation by segmenting the extracted object of step(c) into individual blocks and extracting necessary segment blocks;

e. features extraction from the segment blocks of step(d) by block merging based on feature vectors and matching with histogram based comparison stored in a existing database for feature training;

f. classification of prawn images from the extracted features of step(e) using Neural Networks built by taking the features from existing database.

2. The automatic prawn species recognition system as claimed in claim 1 wherein the said features includes Rostrum, Carapace, Abdomen and Telson.

3. The automatic prawn species recognition system as claimed in claim 1 wherein a modified gradient noise removal algorithm is employed to detect accurate features of Prawn images.

4. The automatic prawn species recognition system as claimed in claim 1 wherein an extraction algorithm is employed to extract foreground images from background.

5. The automatic prawn species recognition system as claimed in claim 1 wherein divide and conquer approach is developed for the block division, identification, modification and merging.

6. The automatic prawn species recognition system as claimed in claim 1 wherein feature extraction algorithms are generated to extract rostrum, carapace, abdomen and telson from the merged structure of a Prawn.

7. The automatic prawn species recognition system as claimed in claim 1 wherein multi-class back propagation neural network model is employed to recognize the prawn species

8. The automatic prawn species recognition system as claimed in claim 1 wherein the classification involves the identification of image prawns; family, Scientific Name, English name and local name.