Research On Shellfish Object Location And Statistics Measurement Based On Deep Learning

Shellfish is a highly sought-after type of seafood.It needs to be statistically measured under different developmental periods to settle down a proper rearing strategy,improving production and quality.Due to the irregular figure in its appearance,shellfish can’t be measured by general Geometric Morphometric.This thesis put forward measurement and statistical methods based on deep learning regarding the location of shellfish to solve the difficulty in measurement.By placing the shellfish object on the specially designed measuring container and taking a photo,the physical size data of the shellfish object,as well as the number of shellfish objects in the image,can be calculated.This method can be used on mobile devices like telephones and laptops.The research work of this thesis is summarized as follows:(1)A container-designed strategy for location measurement is proposed.This thesis applies 4 flat containers with a specified feature identifier to test the feature to identify the location.According to the imaging feature point of the flat container,this test accomplishes the transformation from pixel size to physical size by simple calculation to acquire the Homography Matrix,the congruent relationship between the imaging feature point and the actual container features.This method provides an accurate and reliable measurement for the size statistics of the selected shellfish location.(2)A feature point detection method based on deep learning is proposed.In the above Homography Matrix estimation method,the algorithm needs to detect the feature point of the measuring container in the image.However,the traditional feature point detection method does not perform well in complex conditions.Therefore,this thesis uses a deep learning method to carry out the detection of the container feature point.The experiment results indicate that the feature point position which uses this method is basically the same as the labeled position,proving the validity of the method and the capability of estimating the problems of the Homography Matrix.(3)A measurement and statistics method of shellfish location based on deep learning is proposed.Firstly,using the following feature point detective method mentioned in 1 and2 could obtain the Homography Matrix,which comes from the measurement container flat and physical flat in the image.Next,the machine extracts the shellfish location features through deep learning of the location detective network,and the prediction of shellfish location,width,and height,the number are returned respectively.The shellfish location can be accurately positioned by using a rectangular frame.Finally,the real size of the shellfish location is estimated by using the position frame and the Homography Matrix,taking measures in transforming the rectangular frame of the location.The results show that the detection method could be used to measure shellfish in large quantities for calculating the number of shellfish locations accurately.The difference between the algorithm estimation of size and that of the manual measurement is within 2MM.In conclusion,this thesis proposes a measurement of shellfish location and number statistic method based on deep learning,which has obtained a higher accuracy in the transformation from the pixel size to the actual size and calculates the number of shellfish.Compared with the traditional manual measurement method,this method has three advantages: higher efficiency and faster speed;accomplishes batch measurement,and can be widely used in real shellfish measurement.