A Multi-Layer Feature Fusion Method for Few-Shot Image Classification.

In image classification, few-shot learning deals with recognizing visual categories from a few tagged examples. The degree of expressiveness of the encoded features in this scenario is a crucial question that needs to be addressed in the models being trained. Recent approaches have achieved encouraging results in improving few-shot models in deep learning, but designing a competitive and simple architecture is challenging, especially considering its requirement in many practical applications. This work proposes an improved few-shot model based on a multi-layer feature fusion (FMLF) method. The presented approach includes extended feature extraction and fusion mechanisms in the Convolutional Neural Network (CNN) backbone, as well as an effective metric to compute the divergences in the end. In order to evaluate the proposed method, a challenging visual classification problem, maize crop insect classification with specific pests and beneficial categories, is addressed, serving both as a test of our model and as a means to propose a novel dataset. Experiments were carried out to compare the results with ResNet50, VGG16, and MobileNetv2, used as feature extraction backbones, and the FMLF method demonstrated higher accuracy with fewer parameters. The proposed FMLF method improved accuracy scores by up to 3.62% in one-shot and 2.82% in five-shot classification tasks compared to a traditional backbone, which uses only global image features.

Chasing Flies: The Use of Wingbeat Frequency as a Communication Cue in Calyptrate Flies (Diptera: Calyptratae).

The incidental sound produced by the oscillation of insect wings during flight provides an opportunity for species identification. Calyptrate flies include some of the fastest and most agile flying insects, capable of rapid changes in direction and the fast pursuit of conspecifics. This flight pattern makes the continuous and close recording of their wingbeat frequency difficult and limited to confined specimens. Advances in sound editor and analysis software, however, have made it possible to isolate low amplitude sounds using noise reduction and pitch detection algorithms. To explore differences in wingbeat frequency between genera and sex, 40 specimens of three-day old Sarcophaga crassipalpis, Lucilia sericata, Calliphora dubia, and Musca vetustissima were individually recorded in free flight in a temperature-controlled room. Results showed significant differences in wingbeat frequency between the four species and intersexual differences for each species. Discriminant analysis classifying the three carrion flies resulted in 77.5% classified correctly overall, with the correct classification of 82.5% of S. crassipalpis, 60% of C. dubia, and 90% of L. sericata, when both mean wingbeat frequency and sex were included. Intersexual differences were further demonstrated by male flies showing significantly higher variability than females in three of the species. These observed intergeneric and intersexual differences in wingbeat frequency start the discussion on the use of the metric as a communication signal by this taxon. The success of the methodology demonstrated differences at the genus level and encourages the recording of additional species and the use of wingbeat frequency as an identification tool for these flies.

Resolving the identification of weak-flying insects during flight: a coupling between rigorous data processing and biology.

Bioacoustic methods play an increasingly important role for the detection of insects in a range of surveillance and monitoring programmes.Weak-flying insects evade detection because they do not yield sufficient audio information to capture wingbeat and harmonic frequencies. These inaudible insects often pose a significant threat to food security as pests of key agricultural crops worldwide.Automatic detection of such insects is crucial to the future of crop protection by providing critical information to assess the risk to a crop and the need for preventative measures.We describe an experimental set-up designed to derive audio recordings from a range of weak-flying aphids and beetles using an LED array.A rigorous data processing pipeline was developed to extract meaningful features, linked to morphological characteristics, from the audio and harmonic series for six aphid and two beetle species.An ensemble of over 50 bioacoustic parameters was used to achieve species discrimination with a success rate of 80%. The inclusion of the dominant and fundamental frequencies improved prediction between beetles and aphids because of large differences in wingbeat frequencies.At the species level, error rates were minimized when harmonic features were supplemented by features indicative of differences in species' flight energies.