Shrimp Shape Analysis by a Chord Length Function Based Methodology

Non-Invasive Fish Biometrics for Enhancing Precision and Understanding of Aquaculture Farming through Statistical Morphology Analysis and Machine Learning

“…In the comparison made, it was demonstrated that differences in the average mean absolute error of models with manually obtained characteristics versus those obtained by the image processing algorithm could be 0.5 g for a power model and 0.69 g for the more feature-and environment-consistent MLP model or in relation to the mean dataset weight, 3% and 4.55% ((MAE manual − MAE CLF )/W datasetMean ). Therefore, these results strongly suggest the validity of the feature extraction method developed in [34] and implemented in a computer vision pipeline for biomass estimation in this work. Investigations could be conducted throughout the fattening or development process in aquaculture ponds or research centers, obtaining histograms that represent the shape and size of different specimens (fish and crustaceans) and thus making comparisons and gaining new insights into morphological and allometric development of animals from an approach of representative statistical parameters of shape and size obtained through a signature function.…”

“…This work advances the development of a non-invasive fish biometrics methodology based on the feature extraction algorithm developed by [34] validating its biomass estimation capabilities on fish in conjunction with a machine learning model and a top-placed camera. Three ML models with different complexities were tested and the multilayer perceptron emerged as the most consistent under different features and environmental conditions.…”

“…Section 3.1 describes the data collection hardware and software used for taking the pictures and measuring the lengths, heights, and weights of the tilapias. Section 3.2 describes the applied feature extraction algorithm developed by [34] and the threshold segmentation techniques used for fitting it to the contours of tilapias.…”

“…However, as evidenced in [32], different life stages of a species may require more features for better biomass estimation and for biological and aquaculture studies. Thus, a feature extraction method such as the one in [34], which provides not only size but also shape parameters of any closed contour, could be of help in the quest for better features and standardizing processes for non-invasive biometrics.…”

“…The algorithm used for shape analysis is proposed by [34] and is described as Algorithm 1 (code was added to compute extra shape and size parameters):…”