This　paper　explores　the　identification　methods　of　three　common　citrus　leaf　diseases　that　are　citrus　canker,　citrus　scab　and　citrus　anthracnose　respectively.　The　traditional　machine　learning　algorithms　typically　suffer　from　the　problem　of　low　recognition　accuracy.　In　this　paper,　we　propose　a　method　based　on　deep　convolutional　network　to　solve　this　problem.　The　core　idea　is　to　build　a　7-layer　network　structure,　through　which　the　main　purpose　is　to　extract　rich　features　of　citrus.　These　features　are　better　than　traditional　features　in　identifying　different　categories　of　diseases,　thus　can　improve　the　recognition　accuracy.　We　propose　a　novel　network　which　includes　input　layer,　three　convolutional　layers,　two　fully　connection　layers　and　one　output　layer.　The　convolutional　layer　includes　a　convolutional　operation　and　a　pooling　operation.　The　proposed　method　yields　good　results　in　the　identification　of　citrus　diseases.　In　order　to　show　that　our　results　are　more　accurate　than　those　of　two　other　common　machine　learning　algorithms,　we　conduct　three　sets　of　experiments.　Finally,　experiments　demonstrate　that　the　proposed　method　is　effective　in　identifying　citrus　diseases,　which　provides　effective　technical　support　for　precise　recognition　and　the　prevention　of　citrus　diseases.　©　2019　IEEE.