Pseudostem Weevil Detector (PSWD)

Acoustic pest detection in banana crops using CNNs.

An innovative acoustic monitoring system designed to help farmers detect Pseudostem Weevil (PSW) infestations during the pupa stage. The system analyzes distinctive acoustic traits to provide early intervention insights.

Technical Stack

  • Deep Learning: Convolutional Neural Networks (CNN)
  • Audio Processing: Librosa for Mel spectrogram feature extraction
  • Backend: Flask (Python)
  • Frontend: HTML5, CSS3, Bootstrap

Core Features

  • Acoustic Data Engineering: Curated a dataset of 6,000 field recordings, utilizing Mel spectrogram extraction to transform raw audio into 2D image representations for deep learning.

  • Deep Learning Excellence: Trained a CNN classifier that achieved 99.75% accuracy in detecting early-stage infestations, significantly outperforming traditional manual inspection.

  • Full-Stack Deployment: Engineered an end-to-end web portal on PythonAnywhere, enabling real-time audio uploads and instant inference for field use.

Team & Collaboration

Collaborative project with Annie Treasa Sabu, Anuranjana Rajeev, and Nayana P.

Field photos of Pseudostem Weevil samples collected during data acquisition.