FishTrader: market intelligence platform for profit maximization of fish farmers and fish industry

• Grant number: 22/08224-4
• Support Opportunities: Research Grants - Innovative Research in Small Business - PIPE
• Start date: March 01, 2023
• End date: February 28, 2025
• Field of knowledge: Agronomical Sciences - Fishery Resources and Fishery Engineering - Aquaculture
• Agreement: SEBRAE-SP
• Principal Investigator: Tiago Zanett Albertini
• Grantee: Tiago Zanett Albertini
• Company: Tech Inovações Tecnológicas para a Agropecuária S/A (Filial)
• CNAE: Desenvolvimento de programas de computador sob encomenda
• Associated research grant: 19/27218-2 - FishTrader: market intelligence platform for profit maximization of fish farmers and the fish industry, AP.PIPE
• Associated research grant(s): 22/15576-4 - FishTrader: market intelligence platform for profit maximization of fish farmers and fish industry, AP.PIPE
• Associated scholarship(s): 23/14675-1 - Development of the Data Science and Operational Research solution for the FishTrader Project, BP.TT; 23/13597-7 - Development of machine learning algorithms to test fish biometric parameters, BP.TT; 23/13006-9 - Support in the development of the Data Science and Operational Research solution for the FishTrader Project, BP.TT; + associated scholarships 23/13301-0 - Support in the development of APIs, databases and web and mobile applications for the Fishtrader platform, BP.TT; 23/08684-8 - FishTrader: market information intelligence platform for profit maximization for fish farmers and the seafood industry, BP.TT; 23/07034-0 - Development of APIs, database and web and mobile application for the FishTrader platform, BP.TT; 23/07175-2 - FishTrader: market intelligence platform for profit maximization of fish farmers and fish industry, BP.PIPE - associated scholarships

Abstract

The FishTrader (Processo Fapesp PIPE Fase 1 2019/27218-2) has as its value proposition an intelligent platform that predicts the ideal time for harvest and slaughter fish (weight, size and volume range) based on market information and automatically generated biometrics data. The purpose of this innovation is that fish farms and industries maximize the profitability of their activities through assertive measurements of the Optimal Trading Point (OTP), in a model that optimizes trades in daily processing scale according to production indicators adopted by this market and controlled from automated data on the platform. In PIPE 1, data were collected from tilapia, tambaqui and matrinxã production chain, interviewing 20 fish farms, 6 slaughterhouses, 6 vertical companies (produce, process and sell) and 3 cooperatives. The study of trading rules between them allowed us to understand that the only characteristic considered is the average weight of the fish, obtained manually and by sampling. However, 100% of the slaughterhouses interviewed declare the lack of more accurate weight information, and it is common to obtain lighter fish, a situation that results in lower fillet yield. The fish farmers, on the other hand, value the shortest growing period for lower feed expenses. Thus, for the OTP model, factors such as feed conversion were considered, in which the most efficient batches will be kept longer in fattening for both sectors to obtain maximum gain in profit. In addition to the OTP model, a study of a biometric algorithm that predicts weight using weight and length relationships using stereoscopic cameras was performed. These were able to aggregate depth data, being carried out in four steps: 1) image capture performed with two stereoscopic cameras, the Intel Realsense D435 and the Ailipu ELP-960P2CAM, the latter water-resistant; 2) application of a trained deep learning model to perform the segmentation of fish images under controlled conditions; 3) analysis of each fish segmented by the algorithm, generating biometric information (weight, total length and height); and 4) recording the best predictions in a dataset. Finally, it performed a comparison between the biometric results provided by the algorithm and the manual biometric results, testing the camera system and its predictive capabilities, at four turbidity levels: 3, 20, 60 and 100 Nephelometric Turbidity Unit (NTU). From the collection of manual biometrics data, was built a function capable of relating the total length measurement with the respective tilapia weight. Errors between 7% and 13% were found for values of average total length, and errors between 8% and 26% for values of average weight of the set of fish evaluated. The objective of the current proposal in Phase 2 is to develop the first prototype of the platform in the field, which will be tested in partner fish farms, in the states with the highest national production (SP, PR and RO), with modules 1 and 2. In module 1 will be used biometric data provided by fish farmers; in module 2 will be used fish measurements daily generated by the 3DFish system using artificial intelligence and computer vision. The animal model will be tilapia with an evaluation by sampling, however, individual recognition technology using machine learning, idGigas, will be integrated into FishTrader, from the bone marks on the head of the pirarucu (Arapaima gigas) which has high market value. The information from both modules must supply the same database and be used to define OTP by the performance and profitability curves integrated to an operational research model, using statistical and machine learning algorithms. Some market relevance companies became partners in this project and can be users or assist the technical-business-commercial evolution of the proposal. (AU)