Aabi Zaraat.ai

Abstract

In the realm of research and development, our Final Year Project (FYP) introduces an innovative mobile application tailored for business-to-business (B2B) users. Leveraging smartphone cameras, users can seamlessly identify and classify diverse soil types and fish species found across Pakistan. For soil classification, the app provides comprehensive insights into soil properties, recommends compatible fish species, and details precise ingredient proportions (fish hydrolysate) for optimal soil health. In fish classification, the app offers insights into fish health, freshness, and disease presence, guiding users on ideal combinations for optimal results.

Keywords: Artificial Intelligence, Deep Learning, Computer Vision, Agriculture, Aquaculture, Economical.

Introduction

Recent advancements in Artificial Intelligence (AI), particularly in Deep Learning (DL) and Computer Vision (CV), have paved the way for transformative mobile applications. Our project focuses on the intersection of AI, agriculture, and aquaculture in Pakistan. Agriculture being the backbone of Pakistan's economy, our project aims to bridge the gap between technology and farming.

Imagine a mobile application revolutionizing soil classification and fish assessment using the smartphone's camera. Designed for B2B users, it opens possibilities in Pakistan's agricultural landscape. For soil classification, the app becomes a knowledgeable ally for farmers, offering insights into soil properties, recommending compatible fish species, and providing precise instructions for fish hydrolysate incorporation. In fish classification, the app provides extensive information about fish health, freshness, and diseases, guiding users on optimal combinations.

With this innovative project, we aim to create a farming companion, simplifying agriculture complexities and automating the Fish Hydrolysate procedure. The goal is to promote intelligent and sustainable agricultural practices.

Description (Rationale) Of the Proposed Work

The project addresses the classification of different soil and fish types in Pakistan, optimizing the fish hydrolysate procedure. Using locally sourced datasets from Karachi, the app is finely tuned to the specific soil and fish varieties in the region. The proposed system automates the process via AI, providing users with relevant details after processing images of fish or soil.

Research Objectives

1. Soil Classification: Develop a robust system to identify and classify various soil types in Pakistan.

2. Fish Classification: Implement a comprehensive fish classification module within the app.

3. Data Collection: Gather extensive datasets of soil samples and fish species from the local Karachi market.

4. Algorithm Development: Use deep learning algorithms for soil and fish classification, achieving a classification accuracy of nearly 95%.

5. User Interface: Design an intuitive and user-friendly interface for B2B users.

6. Database Integration: Create a centralized database for storing information about soil types, fish species, and their compatibility.

7. Recommendation System: Implement a system that suggests the most compatible fish species for a given soil type.

8. Health Assessment: Develop a health assessment system for fish, providing detailed information about their condition.

Project Phases and Research Methodology

Project Phases

Phase 1: Comprehensive Literature Review

We conduct a thorough literature survey from various resources to determine state-of-the-art techniques in artificial intelligence and explore practices in agriculture and aquaculture.

Phase 2: Data Collection

Visit local fish markets in Karachi to collect datasets through pictures, extending to soil datasets locally found in Pakistan.

Phase 3: Data-pre-processing (Feature Extraction)

After dataset collection, extract features from fish and soil images, involving cleaning, organizing, and preparing the data for analysis.

Phase 4: Deep Learning Model

Leverage deep learning techniques, transfer learning, and fine-tuning strategies to train the model using prepared datasets, aiming for over 90% accuracy.

Phase 5: Mobile Application Development

Develop a user-friendly app compatible with business-to-business (B2B) use, with the deep learning model handling the classification procedure.

Phase 6: Integration and Testing

Rigorously test the application on real-time local data (Karachi) to ensure functionality, providing valuable feedback for refinement.

Phase 7: Evaluation

Evaluate the model using confusion matrices and other validation techniques. Address shortcomings and implement optimization strategies.

Phase 8: Writing-up and Documentation

Maintain comprehensive documentation of methodologies, data, code, and results throughout the project, serving as the basis for journal articles, conference papers, and project reports.

Research Methodology

Methodology

Utilize deep learning for classification, with an experimental setup for on-site testing, deep learning model training and testing phases, and evaluation criteria for system performance.

Data Collection

Visit various sites in Karachi to collect an image dataset for fish and soil classification, adhering to standardized methods and protocols for accuracy and consistency.

Feature Extraction

Shift focus to feature extraction after data collection, extracting features related to soil composition, texture, nutrient content, fish morphology, coloration, and scale patterns.

Deep Learning Model

Integrate different models (ResNet V0, GoogleNet, EfficientNet) for each stage, configuring models based on their accuracy during testing.

Evaluation

Apply test data with abnormalities to the trained model, assessing performance through confusion matrices and abnormality graphs.