Résumé
Real-time monitoring of hematophagous diptera(such as mosquitoes) populations in the field is a crucial challengeto foresee vaccination campaigns and to restrain potential diseasesspreading. However, current methods heavily rely on costlyDNA extraction which is destructive, costly, time consuming andrequires experts. The contributions of this work are: 1) the usageof a new type of imaging, named Wing Interference Patterns(WIPs), which is non-destructive and easier to produce duringin the field experiments; 2) a deep learning architecture whichis optimized for very low computation cost, memory usage anda short inference time; 3) the use of a dataset of more than50 medically important species of hematophagous diptera withmore than 3000 images of WIPs. With these contributions, wedemonstrate that WIPs are an excellent medium to automaticallyrecognize a large amount of hematophagous diptera species withvery high accuracy and low computational cost convolutionalneural network.