Technology Category
Machine Learning: Hardware-aware MLCognitive Modeling & Cognitive Systems: Neural Spike CodingComputer Vision: Learning & Optimization for CV
Convolutional Neural Network Optimization for Beehive Classification Using Bioacoustic Signals
🤖 AI Summary
This study addresses non-intrusive hive health monitoring by proposing a bee acoustic state classification method leveraging time-frequency representation and model lightweighting. To overcome the limited discriminative capacity of conventional spectrograms, we innovatively adopt cochleagrams as the primary input feature, achieving a 98.31% cross-domain classification accuracy on a public honeybee acoustic dataset. To enable edge deployment, we further integrate channel pruning, INT8 quantization, and knowledge distillation to construct an efficient lightweight CNN: reducing parameters by 91.8% and accelerating inference by 66%. This work constitutes the first systematic validation of cochleagrams’ superiority in honeybee bioacoustic recognition and establishes a practical, low-power acoustic monitoring paradigm for agricultural IoT applications.
Technology Category
Application Category
📝 Abstract
The behavior of honeybees is an important ecological phenomenon not only in terms of honey and beeswax production but also due to the proliferation of flora and fauna around it. The best way to study this significant phenomenon is by non-invasive monitoring of beehives using the sounds produced by various body movements that give out audio signals which can be exploited for various predictions related to the objectives mentioned above. This study investigates the application of Convolutional Neural Networks to classify and monitor different hive states with the help of joint time and frequency image representations such as Spectrogram, Mel-Spectrogram, Smoothed-Spectrogram, and Cochleagram. Our findings indicate that the Cochleagram outperformed all the other representations, achieving an accuracy of 98.31% on unseen data. Furthermore, we employed various strategies including pruning, quantization, and knowledge distillation to optimize the network and prevent any potential issues with model size. With these optimizations, the network size was lowered by 91.8% and the inference time was accelerated by 66%, increasing its suitability for real-time applications. Thus our study emphasizes the significance of using optimization approaches to minimize model size, avoid deployment problems, and expedite inference for real-time application as well as the selection of an appropriate time-frequency representation for optimal performance.
Problem
Research questions and friction points this paper is trying to address.
Classifying beehive states using bioacoustic signals from honeybee movements
Optimizing CNN models through pruning and quantization for real-time deployment
Evaluating time-frequency representations like Cochleagram for audio classification accuracy
Innovation
Methods, ideas, or system contributions that make the work stand out.
CNN optimization with pruning, quantization, distillation
Cochleagram time-frequency representation for classification
Reduced model size by 91.8% for real-time use
🔎 Similar Papers
No similar papers found.
