FESNet: Frequency-Enhanced Saliency Detection Network for Grain Pest Segmentation

SIMPLE SUMMARY: Insect pests cause major nutritional and economic losses in stored grains through their pestilential activities, such as feeding, excretion, and reproduction. Therefore, the detection of grain pests and the estimation of their population density are necessary for taking the proper management initiatives in order to control insect infestation. The popular techniques for the detection of grain pests include probe sampling, acoustic detection, and image recognition, among which the image recognition can provide rapid, economic and accurate solutions for the detection of grain pests. With the development of deep learning, convolutional neural networks (CNN) have been extensively used in image classification and object detection. Nevertheless, the pixel-level segmentation of small pests from the cluttered grain background remains a challenging task in the detection and monitoring of grain pests. Inspired by the observation that humans and birds can find the insects in grains with a glance, we propose a saliency detection model to detect the insects in pixels. Firstly, we construct a dedicated dataset, named GrainPest, with small insect objects in realistic storage scenes. Secondly, frequency clues for both the discrete wavelet transformation (DWT) and the discrete cosine transformation (DCT) are leveraged to enhance the performance of salient object segmentation. Moreover, we design a new receptive field block, aggregating multiscale saliency features to improve the segmentation of small insects. ABSTRACT: As insect infestation is the leading factor accounting for nutritive and economic losses in stored grains, it is important to detect the presence and number of insects for the sake of taking proper control measures. Inspired by the human visual attention mechanism, we propose a U-net-like frequency-enhanced saliency (FESNet) detection model, resulting in the pixelwise segmentation of grain pests. The frequency clues, as well as the spatial information, are leveraged to enhance the detection performance of small insects from the cluttered grain background. Firstly, we collect a dedicated dataset, GrainPest, with pixel-level annotation after analyzing the image attributes of the existing salient object detection datasets. Secondly, we design a FESNet with the discrete wavelet transformation (DWT) and the discrete cosine transformation (DCT), both involved in the traditional convolution layers. As current salient object detection models will reduce the spatial information with pooling operations in the sequence of encoding stages, a special branch of the discrete wavelet transformation (DWT) is connected to the higher stages to capture accurate spatial information for saliency detection. Then, we introduce the discrete cosine transform (DCT) into the backbone bottlenecks to enhance the channel attention with low-frequency information. Moreover, we also propose a new receptive field block (NRFB) to enlarge the receptive fields by aggregating three atrous convolution features. Finally, in the phase of decoding, we use the high-frequency information and aggregated features together to restore the saliency map. Extensive experiments and ablation studies on our dataset, GrainPest, and open dataset, Salient Objects in Clutter (SOC), demonstrate that the proposed model performs favorably against the state-of-the-art model.