Enhancing aphid detection framework based on ORB and convolutional neural networks

Abstract

Methods to detect directly aphids based on convolutional neural networks (CNNs) are unsatisfactory because aphids are small and usually are specially distributed. To enhance aphid detection efficiency, a framework based on oriented FAST and rotated BRIEF (ORB) and CNNs (EADF) is proposed by us to detect aphids in images. Firstly, the key point is to find regions of aphids. Points generated by the ORB algorithm are processed by us to generate suspected aphid areas. Regions are fed into convolutional networks to train the model. Finally, images are detected in blocks with the trained model. In addition, in order to solve the situation that the coordinates are not uniform after the image is segmented, we use a coordinate mapping method to unify the coordinates. We compare current mainstream target detection methods. Experiments indicate that our method has higher accuracy than state-of-the-art two-stage methods that the AP value of RetinaNet with EADF is 0.385 higher than RetinaNet without it and the Cascade-RCNN with EADF is more than without it by 43.3% on value of AP, which demonstrates its competency.

Weakly Supervised Attention-based Models Using Activation Maps for Citrus Mite and Insect Pest Classification

“…We plan to deploy our models on mobile devices and use them in the field. It is noteworthy that new pest benchmarks have ad-dressed tiny regions [6,7,41]. Introducing images with small ROIs may become frequent in pest datasets because pests often represent small areas in uncontrolled field images.…”

“…New works have studied tiny insects. For example, the following works detected and localized aphids in wild conditions [6,7], and counted and classified aphids in controlled environments [41]. Aphids are small but visible, unlike some species of mites, which are unrecognizable without magnification.…”

“…Despite recent abounds of new benchmarks on agronomic pests and diseases [5,6,7], most threats are regional and mite presence is rare. Datasets often contain images from the Internet, such as the Insect Pest (IP102) [5].…”

Métodos fracamente supervisionados para classificação de ácaros na citricultura

“…Outro desafio importante é a inadequação das técnicas existentes de aprendizado profundo para extrair características de áreas muito pequenas durante o decorrer do processo de classificação de imagens [102,139,190]. As arquiteturas são normalmente propostas para tarefas gerais e de ponta a ponta -treinadas de forma única e monolítica -, pois os extratores de características, na maior parte das redes, são construídos com downsampling (reduções), que descartam características mais finas em detrimento de dados semanticamente mais gerais.…”

“…Lins et al [110] também trabalharam com pulgões e apresentaram uma ferramenta para automatizar sua contagem e classificação usando métodos de processamento de imagem, visão computacional e DNNs para classificar características locais. Pei et al [139] argumentaram que métodos para detectar pulgões baseados em DNNs seriam insatisfatórios porque os pulgões seriam pequenos e, em geral, espacialmente distribuídos. Eles utilizaram descritores locais binários para encontrar regiões que possivelmente teriam pulgões e produzir localizações aplicando redes convolucionais para essas regiões.…”