鍝庯紝浣犲彂鐜版病锛熺幇鍦ㄨ繛鎵嬫満鐩稿唽閮借兘鑷姩璇嗗埆浜鸿劯浜嗭紝鍋滆溅鍦洪椄鏈虹湅涓€鐪艰溅鐗屽氨鎶潌锛岃繖浜?鑱槑"鐨勬搷浣滆儗鍚庡晩锛屽叾瀹為兘钘忕潃涓€涓彨CNN鐨勬妧鏈€備粖澶╁挶浠氨鏉ュ敔鍞犺繖涓璁＄畻鏈?闀跨溂鐫?鐨勭粷娲烩€斺€斿嵎绉缁忕綉缁滐紒

锛堝仠椤夸笁绉掞級绛夌瓑锛屼綘璇ヤ笉浼氫互涓烘垜瑕佽鏂伴椈鐢佃鍙板惂锛熷搱鍝堝埆鎱岋紝鎴戣鐨凜NN鏄疌onvolutional Neural Network锛屼笓闂ㄥ鐞嗗浘鍍忔暟鎹殑瓒呯骇宸ュ叿銆傚噯澶囧ソ浣犵殑灏忔澘鍑筹紝鍜变滑浠庨浂寮€濮嬬洏瀹冿紒

涓€銆丆NN涓哄暐姣旇倝鐪艰繕鍘夊锛?/h3>

锛堟暡榛戞澘锛夊厛瑙ｅ喅鏈€鏍规湰鐨勯棶棰橈細鍑暐璁╄绠楁満鐪嬪浘鍍忥紵浣犳兂鍟婏紝鍜变滑浜虹溂鑳界灛闂磋鍑虹尗鐙楋紝浣嗚绠楁満鐪嬪埌鐨勫叾瀹炴槸涓€鍫嗘暟瀛楃煩闃点€傝繖鏃跺€機NN灏卞儚缁欐満鍣ㄦ埓浜嗗壇鏅鸿兘鐪奸暅鈥斺€旇繕鏄嚜甯﹀垎鏋愬姛鑳界殑閭ｇ锛?/p>

涓句釜鏍楀瓙馃尠锛氫綘鏁欎笁宀佸▋璁よ嫻鏋滐紝浼氭寚鐫€褰㈢姸棰滆壊璇寸壒寰佸鍚э紵CNN骞茬殑浜嬪樊涓嶅锛?/p>

1. 鈥?strong>鈥嬭疆寤撴壂鎻忊€?/strong>鈥嬶細鍏堢湅杈圭紭鏄渾鐨勮繕鏄柟鐨?/li>
2. 鈥?strong>鈥嬪眬閮ㄧ壒寰佲€?/strong>鈥嬶細瑙傚療鏈夋病鏈夋灉钂傛垨鍙跺瓙
3. 鈥?strong>鈥嬫暣浣撳垽鏂€?/strong>鈥嬶細缁煎悎鎵€鏈変俊鎭‘璁ゆ槸鑻规灉

锛堢獊鐒舵媿澶ц吙锛夊浜嗭紒杩欏氨鍍忕牬妗堟椂鍏堟壘鎸囩汗锛屽啀鐪嬫瘺鍙戯紝鏈€鍚庣患鍚堣瘉鎹攣瀹氬珜鐤戜汉鍢涳紒

---

浜屻€丆NN鐨勪笁澶х湅瀹舵湰棰?/h3>

1. 鍗风Н灞傦細鑷甫鏀惧ぇ闀滅殑渚﹀療鍏?/h4>

鎯宠薄鐢ㄦ斁澶ч暅婊戣繃鍥剧墖锛屾瘡娆″彧鐪嬩竴灏忓潡鍖哄煙銆傝繖涓搷浣滀笓涓氱偣鍙€?strong>鈥嬪眬閮ㄦ劅鐭モ€?/strong>鈥嬶紝濂藉澶т簡鍘讳簡锛?/p>

• 姣斿叏杩炴帴绁炵粡缃戠粶鐪?0%鐨勮绠楅噺
• 鑳芥崟鎹夊钩绉讳笉鍙樼殑鍥炬锛堟瘮濡傜尗鑰虫湹鍦ㄥ乏鍦ㄥ彸閮借寰楀嚭锛?/li>
• 鑷姩鎻愬彇杈圭紭銆佺汗鐞嗙瓑搴曞眰鐗瑰緛

锛堢獊鐒跺彂闂級閭ｄ负鍟ヤ笉鐢ㄨ€佸紡鍥惧儚澶勭悊绠楁硶锛熼棶寰楀ソ锛佷紶缁熸柟娉曞氨鍍忓浐瀹氭祦绋嬬殑娴佹按绾匡紝鑰孋NN鐨勫嵎绉牳鏄兘鑷繁瀛︿範鍗囩骇鐨勶紒

2. 姹犲寲灞傦細鎶撳ぇ鏀惧皬鐨勬櫤鎱?/h4>

缁忓巻杩囨湡鏈垝閲嶇偣鍚楋紵姹犲寲灞傚氨鏄共杩欎釜鐨勶細

• 鈥?strong>鈥嬫渶澶ф睜鍖栤€?/strong>鈥嬶細鍙鍖哄煙鏈€鏄庢樉鐗瑰緛
• 鈥?strong>鈥嬪钩鍧囨睜鍖栤€?/strong>鈥嬶細璁版暣浣撹秼鍔?/li>
• 鍏稿瀷鎿嶄綔锛氭妸4x4鍖哄煙鍘嬬缉鎴?x2锛屾棦淇濈暀鍏抽敭淇℃伅鍙堥槻杩囨嫙鍚?/li>

锛堟暡妗屽瓙寮鸿皟锛夎繖鎷涚壒鍒浠樺浘鍍忓彉褰㈣秴绠＄敤锛佸氨绠楀浘鐗囪鏃嬭浆缂╂斁锛屼富瑕佺壒寰佽繕鍦ㄦ睜鍖栧眰閲岀暀鐫€鍛€?/p>

3. 鍏ㄨ繛鎺ュ眰锛氭渶缁堝喅绛栬€?/h4>

鍒拌繖鍎垮氨璇ユ媿鏉夸簡锛佹妸鍓嶉潰鎻愬彇鐨勭壒寰侊細

• 鍍忎箰楂樼Н鏈ㄤ竴鏍锋嫾鎺?/li>
• 閫氳繃鏉冮噸鍒嗛厤閲嶈鎬?/li>
• 杈撳嚭姒傜巼鍒ゆ柇缁撴灉

锛堢獊鐒舵彃鍏ュ喎鐭ヨ瘑锛?012骞碅lexNet鍦↖mageNet姣旇禌涓€楦ｆ儕浜猴紝灏辨槸闈犺繖涓変釜娉曞疂鎶婇敊璇巼浠?6%骞插埌16.4%锛岀洿鎺ユ巰璧锋繁搴﹀涔犻潻鍛斤紒

---

涓夈€佹墜鎶婃墜鎼缓绗竴涓狢NN妯″瀷

鍜变滑鐢≒ython鍜孠eras婕旂ず涓瀬绠€鐗堬紝鍏堟悶鎳傛祦绋嬶細

python澶嶅埗
from keras.models import Sequential
from keras.layers import Conv2D, MaxPooling2D, Flatten, Dense

model = Sequential()
# 绗竴缁勬斁澶ч暅
model.add(Conv2D(32, (3,3), activation='relu', input_shape=(64,64,3)))
# 鍒掗噸鐐?/span>
model.add(MaxPooling2D(pool_size=(2,2)))
# 鍐嶆潵涓€缁勯珮绾ф斁澶ч暅
model.add(Conv2D(64, (3,3), activation='relu'))
# 鍐嶅垝閲嶇偣
model.add(MaxPooling2D(pool_size=(2,2)))
# 鎷间箰楂?/span>
model.add(Flatten())
# 鏈€缁堝喅绛?/span>
model.add(Dense(units=128, activation='relu'))
model.add(Dense(units=1, activation='sigmoid'))

锛堟彃鍏ヨ皟璇曟晠浜嬶級璁板緱鎴戠涓€娆¤繍琛屾椂锛屾妸杈撳叆灏哄鎼為敊浜嗭紝缁撴灉妯″瀷姝绘椿涓嶅伐浣?..鎵€浠ユ柊鎵嬪崈涓囨敞鎰廼nput_shape瑕佸拰鏁版嵁闆嗕竴鑷村晩锛?/p>

鍥涖€侀伩寮€鏂颁汉蹇呰俯鐨?涓ぇ鍧?/h3>

1. 鈥?strong>鈥嬫暟鎹笉澶熻繕纭垰鈥?/strong>鈥嬶細寤鸿灏戜簬1涓囧紶鍥炬椂鐢ㄨ縼绉诲涔?/li>
2. 鈥?strong>鈥嬫棤鑴戝爢鍗风Н灞傗€?/strong>鈥嬶細灞傛暟澶鍙嶈€屼細姊害娑堝け锛堜翰娴?-5灞傚鐢級
3. 鈥?strong>鈥嬪拷瑙嗘暟鎹寮衡€?/strong>鈥嬶細鏃嬭浆/闀滃儚/瑁佸壀鑳芥彁鍗?0%鏁堟灉
4. 鈥?strong>鈥嬪繕璁版鍒欏寲鈥?/strong>鈥嬶細Dropout鍜孡2姝ｅ垯鍖栨槸闃茶繃鎷熷悎绁炲櫒
5. 鈥?strong>鈥嬬洸鐩皟鍙傗€?/strong>鈥嬶細鍏堢敤缃戞牸鎼滅储纭畾澶ц嚧鑼冨洿鍐嶅井璋?/li>

锛堢獊鐒舵彁楂橀煶閲忥級閲嶇偣鏉ヤ簡锛佸鐞嗗尰鐤楀奖鍍忔椂锛屾暟鎹寮哄崈涓囧埆鐢ㄩ暅鍍忕炕杞€斺€斾綘鎬讳笉甯屾湜CT鐗囧乏鍙抽鍊掑惂锛?/p>

---

浜斻€佷釜浜鸿娉粡楠岃皥

鎼炰簡涓夊勾CV椤圭洰锛岃鐐规暀绉戜功涓嶄細鍐欑殑锛?/p>

• 鍒鍑嗙‘鐜囧拷鎮狅紝鍖荤枟椤圭洰鏇寸湅閲嶅彫鍥炵巼
• 宸ヤ笟閮ㄧ讲瑕佽€冭檻鎺ㄧ悊閫熷害锛孧obileNet姣擵GG瀹炵敤
• 閬囧埌鐗瑰緛娣锋穯鏃讹紙姣斿姗樼尗鍜岃€佽檸锛夛紝璇曡瘯娉ㄦ剰鍔涙満鍒?/li>
• 鏈€鏂拌秼鍔挎槸CNN+Transformer娣峰悎鏋舵瀯锛堜絾鏂版墜鍏堟墦濂藉熀纭€锛?/li>

锛堝帇浣庡０闊筹級鍋峰伔璇达紝Kaggle姣旇禌閲岋紝闆嗘垚澶氫釜CNN妯″瀷缁忓父鑳藉啿杩涘墠鎺掋€備笉杩囩湡瀹為」鐩鑰冭檻鎴愭湰锛屽埆涓轰簡1%鐨勬彁鍗囧鑺变笁鍊嶇畻鍔?..

鍏€丆NN杩樿兘杩欎箞鐜╋紵锛?/h3>

浣犱互涓哄畠鍙兘澶勭悊鍥惧儚锛熸牸灞€鎵撳紑锛?/p>

• 瑙嗛鍒嗘瀽锛氭妸杩炵画甯у綋绔嬩綋鏁版嵁
• 鏂囨湰鍒嗙被锛氭妸鏂囧瓧杞垚璇嶅悜閲忕煩闃?/li>
• 璇煶璇嗗埆锛氬０璋卞浘涔熸槸浜岀淮鏁版嵁
• 鐢氳嚦棰勬祴鑲＄エ锛堣櫧鐒舵晥鏋滆浠佽鏅猴級

锛堢獊鐒惰嚜鍢诧級鍘诲勾鐢–NN鍋氬績鐢靛浘鍒嗙被锛岀粨鏋滄妸鎴块ⅳ鍜屾甯稿績璺虫悶娣蜂簡...鍚庢潵鍙戠幇鏄暟鎹爣娉ㄦ湁闂銆傛墍浠ュ晩锛屾暟鎹川閲忔瘮妯″瀷鏇撮噸瑕侊紒

---

鐪嬪畬鏄笉鏄寰桟NN涔熸病閭ｄ箞鐜勪箮锛熻鐧戒簡灏辨槸妯′豢浜虹被瑙嗚绯荤粺鐨勫伐浣滄柟寮忋€傛渶鍚庨€佸ぇ瀹朵袱鍙ヨ瘽锛氬叾涓€锛岃窇閫氱涓€涓ā鍨嬪悗鍗冧竾鍒槡鐟燂紝鍙兘鍙槸鏁版嵁闆嗗お绠€鍗曪紱鍏朵簩锛岄亣鍒扮摱棰堟椂澶氭兂鎯斥€斺€旇鏄汉绫讳笓瀹朵細鎬庝箞澶勭悊杩欎釜闂锛?/p>

锛堟枃绔犳垱鐒惰€屾锛屽儚绐佺劧鎺ュ埌鐢佃瘽鍖嗗繖缁撴潫锛夊搸鍛€瀹㈡埛鎵炬垜鏀规ā鍨嬩簡锛屼笅娆″啀鑱婂叾浠栫粏鑺傦紒璁板緱鍔ㄦ墜瀹炶返鎵嶆槸纭亾鐞嗭紝鍏夌湅涓嶅姩姘歌繙瀛︿笉浼氬搱~